db/pgbackrest
1
0
Fork 0
mirror of https://github.com/pgbackrest/pgbackrest.git synced 2025-04-21 11:57:01 +02:00
Code Issues Releases Activity
pgbackrest/test/src/module/command/backupTest.c

3277 lines
170 KiB
C
Raw Normal View History

The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
/***********************************************************************************************************************************
Test Backup Command
***********************************************************************************************************************************/
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
#include "command/stanza/create.h"
#include "command/stanza/upgrade.h"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
#include "common/crypto/hash.h"
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
#include "common/io/bufferWrite.h"
Use an extension to denote vendorized code. Vendorized code is copied from another project when a library is not available and a git subproject won't work. Currently all the vendorized code is copied from PostgreSQL but it makes sense to have a more general mechanism for indicating vendorized code. The .vendor extension will be used to denote vendorized code in the same way that .auto is used to denote auto-generated code.
2020-05-18 19:11:26 -04:00
#include "postgres/interface/static.vendor.h"
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
#include "storage/helper.h"
#include "storage/posix/storage.h"
#include "common/harnessConfig.h"
Move PostgreSQL version interface test functions to a test harness. Some version interface test functions were integrated into the core code because they relied on the PostgreSQL versioned interface. Even though they were compiled out for production builds they cluttered the core code and made it harder to determine what was required by core. Create a PostgreSQL version interface in a test harness to contain these functions. This does require some duplication but the cleaner core code seems a good tradeoff. It is possible for some of this code to be auto-generated but since it is only updated once per year the matter is not pressing.
2021-05-17 07:20:28 -04:00
#include "common/harnessPostgres.h"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
#include "common/harnessPq.h"
Convert page checksum filter result to a pack. The pack is both more compact and more efficient than a variant. Also aggregate the page error info in the main process rather than in the filter to allow additional LSN filtering, to be added in a future commit.
2021-09-24 17:40:31 -04:00
#include "common/harnessPack.h"
Add local process shim. Run the local process inside a forked child process instead of exec'ing it. This allows coverage to accumulate in the local process rather than needing to test the local protocol functions directly, resulting in better end-to-end testing and less test duplication. Another advantage is that the pgbackrest binary does not need to be built for the test. The backup, restore, and verify command tests have been updated to use the new shim for coverage.
2021-05-21 12:45:00 -04:00
#include "common/harnessProtocol.h"
Add backup delta unit test. This test was added to take the place of another test, which turned out not to be workable. Even so, it adds coverages at little cost so it seems worth keeping.
2021-03-11 14:40:14 -05:00
#include "common/harnessStorage.h"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
/***********************************************************************************************************************************
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
Get a list of all files in the backup and a redacted version of the manifest that can be tested against a static string
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
***********************************************************************************************************************************/
typedef struct TestBackupValidateCallbackData
{
const Storage *storage; // Storage object when needed (e.g. fileCompressed = true)
const String *path; // Subpath when storage is specified
Pack manifest file structs to save memory. Manifests with a very large number of files can use a considerable amount of memory. There are a lot of zeroes in the data so it can be stored more efficiently by using base-128 varint encoding for the integers and storing the strings in the same allocation. The downside is that the data needs to be unpacked in order to be used, but in most cases this seems fast enough (about 10% slower than before) except for saving the manifest, which is 10% slower up to 10 million files and then gets about 5x slower by 100 million (two minutes on my M1 Mac). Profiling does not show this slowdown so I wonder if this is related to the change in memory layout. Curiously, the function that increased most was jsonFromStrInternal(), which was not modified. That gives more weight to the idea that there is some kind of memory issue going on here and one hopes that servers would be less affected. Either way, they largest use cases we have seen are for about 6 million files so if we can improve that case I believe we will be better off. Further analysis showed that most of the time was taken up writing the size and timestamp fields, which makes almost no sense. The same amount of time was used if they were hard-coded to 0, which points to some odd memory issue on the M1 architecture. This change has been planned for a while, but the particular impetus at this time is that small file support requires additional fields that would increase manifest memory usage by about 20%, even if the feature is not used. Note that the Pack code has been updated to use the new varint encoder, but the decoder remains separate because it needs to fetch one byte at a time.
2022-01-21 17:05:07 -05:00
Manifest *manifest; // Manifest to check for files/links/paths
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
const ManifestData *manifestData; // Manifest data
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
String *content; // String where content should be added
} TestBackupValidateCallbackData;
Add missing static keywords in test modules.
2021-10-18 12:22:48 -04:00
static void
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
testBackupValidateCallback(void *callbackData, const StorageInfo *info)
{
TestBackupValidateCallbackData *data = callbackData;
// Don't include . when it is a path (we'll still include it when it is a link so we can see the destination)
if (info->type == storageTypePath && strEq(info->name, DOT_STR))
return;
// Don't include backup.manifest or copy. We'll test that they are present elsewhere
if (info->type == storageTypeFile &&
(strEqZ(info->name, BACKUP_MANIFEST_FILE) || strEqZ(info->name, BACKUP_MANIFEST_FILE INFO_COPY_EXT)))
return;
switch (info->type)
{
case storageTypeFile:
{
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
// Test mode, user, group. These values are not in the manifest but we know what they should be based on the default
// mode and current user/group.
// ---------------------------------------------------------------------------------------------------------------------
if (info->mode != 0640)
THROW_FMT(AssertError, "'%s' mode is not 0640", strZ(info->name));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
if (!strEq(info->user, TEST_USER_STR))
THROW_FMT(AssertError, "'%s' user should be '" TEST_USER "'", strZ(info->name));
if (!strEq(info->group, TEST_GROUP_STR))
THROW_FMT(AssertError, "'%s' group should be '" TEST_GROUP "'", strZ(info->name));
// Build file list (needed because bundles can contain multiple files)
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
// ---------------------------------------------------------------------------------------------------------------------
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
List *const fileList = lstNewP(sizeof(ManifestFilePack **));
bool bundle = strBeginsWithZ(info->name, "bundle/");
Add infrastructure for multiple compression type support. Add compress-type option and deprecate compress option. Since the compress option is boolean it won't work with multiple compression types. Add logic to cfgLoadUpdateOption() to update compress-type if it is not set directly. The compress option should no longer be referenced outside the cfgLoadUpdateOption() function. Add common/compress/helper module to contain interface functions that work with multiple compression types. Code outside this module should no longer call specific compression drivers, though it may be OK to reference a specific compression type using the new interface (e.g., saving backup history files in gz format). Unit tests only test compression using the gz format because other formats may not be available in all builds. It is the job of integration tests to exercise all compression types. Additional compression types will be added in future commits.
2020-03-06 14:41:03 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
if (bundle)
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
{
Add cvtZSubNTo*() functions. These functions allow conversion from substrings without needing to create a String or a temporary buffer. httpDateToTime() no longer requires a temp mem context. Also improve handling of month search to avoid an allocation. httpUriDecode() no longer requires a temp mem context. jsonReadStr() no longer requires a temp mem context. pgLsnFromWalSegment() no longer requires a temp mem context. pgVersionFromStr() no longer requires a temp mem context. Also do a bit of refactoring. storageGcsCvtTime() no longer leaks six Strings per call. storageS3CvtTime() no longer leaks six Strings per call.
2022-04-28 09:50:23 -04:00
const uint64_t bundleId = cvtZToUInt64(strZ(info->name) + sizeof("bundle"));
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
for (unsigned int fileIdx = 0; fileIdx < manifestFileTotal(data->manifest); fileIdx++)
{
ManifestFilePack **const filePack = lstGet(data->manifest->pub.fileList, fileIdx);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
if (manifestFileUnpack(data->manifest, *filePack).bundleId == bundleId)
lstAdd(fileList, &filePack);
}
}
else
{
const String *manifestName = info->name;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
if (data->manifestData->backupOptionCompressType != compressTypeNone)
{
manifestName = strSubN(
info->name, 0, strSize(info->name) - strSize(compressExtStr(data->manifestData->backupOptionCompressType)));
}
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
ManifestFilePack **const filePack = manifestFilePackFindInternal(data->manifest, manifestName);
lstAdd(fileList, &filePack);
}
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
// Check files
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
// ---------------------------------------------------------------------------------------------------------------------
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
for (unsigned int fileIdx = 0; fileIdx < lstSize(fileList); fileIdx++)
{
ManifestFilePack **const filePack = *(ManifestFilePack ***)lstGet(fileList, fileIdx);
ManifestFile file = manifestFileUnpack(data->manifest, *filePack);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
if (bundle)
strCatFmt(data->content, "%s/%s {file", strZ(info->name), strZ(file.name));
else
strCatFmt(data->content, "%s {file", strZ(info->name));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
// Calculate checksum/size and decompress if needed
// -----------------------------------------------------------------------------------------------------------------
StorageRead *read = storageNewReadP(
data->storage, strNewFmt("%s/%s", strZ(data->path), strZ(info->name)), .offset = file.bundleOffset,
.limit = VARUINT64(file.sizeRepo));
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
if (data->manifestData->backupOptionCompressType != compressTypeNone)
{
ioFilterGroupAdd(
ioReadFilterGroup(storageReadIo(read)), decompressFilter(data->manifestData->backupOptionCompressType));
}
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
Use StringId for type in cryptoHash() and cryptoHmacOne(). This brings hash types in line with cipher types (i.e. a StringId enum) and allows removal of some Strings.
2022-05-25 15:27:53 -04:00
ioFilterGroupAdd(ioReadFilterGroup(storageReadIo(read)), cryptoHashNew(hashTypeSha1));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
uint64_t size = bufUsed(storageGetP(read));
const String *checksum = pckReadStrP(
ioFilterGroupResultP(ioReadFilterGroup(storageReadIo(read)), CRYPTO_HASH_FILTER_TYPE));
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
strCatFmt(data->content, ", s=%" PRIu64, size);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
if (!strEqZ(checksum, file.checksumSha1))
THROW_FMT(AssertError, "'%s' checksum does match manifest", strZ(file.name));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
// Test size and repo-size. If compressed then set the repo-size to size so it will not be in test output. Even the
// same compression algorithm can give slightly different results based on the version so repo-size is not
// deterministic for compression.
// -----------------------------------------------------------------------------------------------------------------
if (size != file.size)
THROW_FMT(AssertError, "'%s' size does match manifest", strZ(file.name));
// Repo size can only be compared to file size when not bundled
if (!bundle)
{
if (info->size != file.sizeRepo)
THROW_FMT(AssertError, "'%s' repo size does match manifest", strZ(file.name));
}
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
if (data->manifestData->backupOptionCompressType != compressTypeNone)
file.sizeRepo = file.size;
// Bundle id/offset are too noisy so remove them. They are checked size/checksum and listed with the files.
// -----------------------------------------------------------------------------------------------------------------
file.bundleId = 0;
file.bundleOffset = 0;
// pg_control and WAL headers have different checksums depending on cpu architecture so remove the checksum from the
// test output.
// -----------------------------------------------------------------------------------------------------------------
if (strEqZ(file.name, MANIFEST_TARGET_PGDATA "/" PG_PATH_GLOBAL "/" PG_FILE_PGCONTROL) ||
strBeginsWith(
file.name, strNewFmt(MANIFEST_TARGET_PGDATA "/%s/", strZ(pgWalPath(data->manifestData->pgVersion)))))
{
file.checksumSha1[0] = '\0';
}
strCatZ(data->content, "}\n");
// Update changes to manifest file
manifestFilePackUpdate(data->manifest, filePack, &file);
}
Pack manifest file structs to save memory. Manifests with a very large number of files can use a considerable amount of memory. There are a lot of zeroes in the data so it can be stored more efficiently by using base-128 varint encoding for the integers and storing the strings in the same allocation. The downside is that the data needs to be unpacked in order to be used, but in most cases this seems fast enough (about 10% slower than before) except for saving the manifest, which is 10% slower up to 10 million files and then gets about 5x slower by 100 million (two minutes on my M1 Mac). Profiling does not show this slowdown so I wonder if this is related to the change in memory layout. Curiously, the function that increased most was jsonFromStrInternal(), which was not modified. That gives more weight to the idea that there is some kind of memory issue going on here and one hopes that servers would be less affected. Either way, they largest use cases we have seen are for about 6 million files so if we can improve that case I believe we will be better off. Further analysis showed that most of the time was taken up writing the size and timestamp fields, which makes almost no sense. The same amount of time was used if they were hard-coded to 0, which points to some odd memory issue on the M1 architecture. This change has been planned for a while, but the particular impetus at this time is that small file support requires additional fields that would increase manifest memory usage by about 20%, even if the feature is not used. Note that the Pack code has been updated to use the new varint encoder, but the decoder remains separate because it needs to fetch one byte at a time.
2022-01-21 17:05:07 -05:00
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
break;
}
case storageTypeLink:
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
strCatFmt(data->content, "%s {link, d=%s}\n", strZ(info->name), strZ(info->linkDestination));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
break;
case storageTypePath:
{
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
strCatFmt(data->content, "%s {path", strZ(info->name));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Check against the manifest
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
// ---------------------------------------------------------------------------------------------------------------------
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
if (!strEq(info->name, STRDEF("bundle")))
manifestPathFind(data->manifest, info->name);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
// Test mode, user, group. These values are not in the manifest but we know what they should be based on the default
// mode and current user/group.
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
if (info->mode != 0750)
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
THROW_FMT(AssertError, "'%s' mode is not 00750", strZ(info->name));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
if (!strEq(info->user, TEST_USER_STR))
THROW_FMT(AssertError, "'%s' user should be '" TEST_USER "'", strZ(info->name));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
if (!strEq(info->group, TEST_GROUP_STR))
THROW_FMT(AssertError, "'%s' group should be '" TEST_GROUP "'", strZ(info->name));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
strCatZ(data->content, "}\n");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
break;
}
case storageTypeSpecial:
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
THROW_FMT(AssertError, "unexpected special file '%s'", strZ(info->name));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
}
static String *
testBackupValidate(const Storage *storage, const String *path)
{
FUNCTION_HARNESS_BEGIN();
FUNCTION_HARNESS_PARAM(STORAGE, storage);
FUNCTION_HARNESS_PARAM(STRING, path);
FUNCTION_HARNESS_END();
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
ASSERT(storage != NULL);
ASSERT(path != NULL);
Rename strNew() to strNewZ() and add parameter-less strNew(). Replace all instances of strNew("") with strNew() and use strNewZ() for non-empty zero-terminated strings. Besides saving a useless parameter, this will allow smarter memory allocation in a future commit by signaling intent, in general, to append or not. In the tests use STRDEF() or VARSTRDEF() where more appropriate rather than blindly replacing with strNewZ(). Also replace strLstAdd() with strLstAddZ() where appropriate for the same reason.
2021-05-21 17:36:43 -04:00
String *result = strNew();
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
MEM_CONTEXT_TEMP_BEGIN()
{
// Build a list of files in the backup path and verify against the manifest
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
Manifest *manifest = manifestLoadFile(storage, strNewFmt("%s/" BACKUP_MANIFEST_FILE, strZ(path)), cipherTypeNone, NULL);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TestBackupValidateCallbackData callbackData =
{
.storage = storage,
.path = path,
.content = result,
.manifest = manifest,
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
.manifestData = manifestData(manifest),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
};
storageInfoListP(storage, path, testBackupValidateCallback, &callbackData, .recurse = true, .sortOrder = sortOrderAsc);
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
// Make sure both backup.manifest files exist since we skipped them in the callback above
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
if (!storageExistsP(storage, strNewFmt("%s/" BACKUP_MANIFEST_FILE, strZ(path))))
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
THROW(AssertError, BACKUP_MANIFEST_FILE " is missing");
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
if (!storageExistsP(storage, strNewFmt("%s/" BACKUP_MANIFEST_FILE INFO_COPY_EXT, strZ(path))))
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
THROW(AssertError, BACKUP_MANIFEST_FILE INFO_COPY_EXT " is missing");
// Output the manifest to a string and exclude sections that don't need validation. Note that each of these sections should
// be considered from automatic validation but adding them to the output will make the tests too noisy. One good technique
// would be to remove it from the output only after validation so new values will cause changes in the output.
// -------------------------------------------------------------------------------------------------------------------------
Buffer *manifestSaveBuffer = bufNew(0);
manifestSave(manifest, ioBufferWriteNew(manifestSaveBuffer));
Rename strNew() to strNewZ() and add parameter-less strNew(). Replace all instances of strNew("") with strNew() and use strNewZ() for non-empty zero-terminated strings. Besides saving a useless parameter, this will allow smarter memory allocation in a future commit by signaling intent, in general, to append or not. In the tests use STRDEF() or VARSTRDEF() where more appropriate rather than blindly replacing with strNewZ(). Also replace strLstAdd() with strLstAddZ() where appropriate for the same reason.
2021-05-21 17:36:43 -04:00
String *manifestEdit = strNew();
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
StringList *manifestLine = strLstNewSplitZ(strTrim(strNewBuf(manifestSaveBuffer)), "\n");
bool bSkipSection = false;
for (unsigned int lineIdx = 0; lineIdx < strLstSize(manifestLine); lineIdx++)
{
const String *line = strTrim(strLstGet(manifestLine, lineIdx));
if (strChr(line, '[') == 0)
{
const String *section = strSubN(line, 1, strSize(line) - 2);
Improve JSON handling. Previously read/writing JSON required parsing/render via a variant, which add many more memory allocations and loops. Instead allow JSON to be read/written serially to improve performance and simplify the code. This also allows us to get rid of many String and Variant constant which are no longer required. The goal is to be able to read/write very large (e.g. gigabyte manifest) JSON structures, which would not be practical with the current code. Note that external JSON (GCS, S3, etc) is still handled using variants. Converting these will require more consideration about key ordering since it cannot be guaranteed as in our own formats.
2022-04-25 09:06:26 -04:00
if (strEqZ(section, INFO_SECTION_BACKREST) ||
strEqZ(section, MANIFEST_SECTION_BACKUP) ||
strEqZ(section, MANIFEST_SECTION_BACKUP_DB) ||
strEqZ(section, MANIFEST_SECTION_BACKUP_OPTION) ||
strEqZ(section, MANIFEST_SECTION_DB) ||
strEqZ(section, MANIFEST_SECTION_TARGET_FILE_DEFAULT) ||
strEqZ(section, MANIFEST_SECTION_TARGET_LINK_DEFAULT) ||
strEqZ(section, MANIFEST_SECTION_TARGET_PATH_DEFAULT))
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
{
bSkipSection = true;
}
else
bSkipSection = false;
}
if (!bSkipSection)
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
strCatFmt(manifestEdit, "%s\n", strZ(line));
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
}
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
strCatFmt(result, "--------\n%s\n", strZ(strTrim(manifestEdit)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
MEM_CONTEXT_TEMP_END();
Rename FUNCTION_HARNESS_RESULT*() macros to FUNCTION_HARNESS_RETURN*(). When the FUNCTION_*_RESULT*() macros were renamed to FUNCTION_*_RETURN_*() in the core code the test harness macros were missed. Update them to make the naming consistent.
2021-03-10 18:42:22 -05:00
FUNCTION_HARNESS_RETURN(STRING, result);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
/***********************************************************************************************************************************
Generate pq scripts for versions of PostgreSQL
***********************************************************************************************************************************/
typedef struct TestBackupPqScriptParam
{
VAR_PARAM_HEADER;
bool startFast;
bool backupStandby;
bool errorAfterStart;
bool noWal; // Don't write test WAL segments
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
bool noPriorWal; // Don't write prior test WAL segments
bool noArchiveCheck; // Do not check archive
Add infrastructure for multiple compression type support. Add compress-type option and deprecate compress option. Since the compress option is boolean it won't work with multiple compression types. Add logic to cfgLoadUpdateOption() to update compress-type if it is not set directly. The compress option should no longer be referenced outside the cfgLoadUpdateOption() function. Add common/compress/helper module to contain interface functions that work with multiple compression types. Code outside this module should no longer call specific compression drivers, though it may be OK to reference a specific compression type using the new interface (e.g., saving backup history files in gz format). Unit tests only test compression using the gz format because other formats may not be available in all builds. It is the job of integration tests to exercise all compression types. Additional compression types will be added in future commits.
2020-03-06 14:41:03 -05:00
CompressType walCompressType; // Compress type for the archive files
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
unsigned int walTotal; // Total WAL to write
Fix error in timeline conversion. The timeline is required to verify WAL segments in the archive after a backup. The conversion was performed base 10 instead of 16, which led to errors when the timeline was ≥ 0xA.
2020-01-21 10:29:46 -07:00
unsigned int timeline; // Timeline to use for WAL files
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
} TestBackupPqScriptParam;
Remove extra spaces before macro continuation.
2021-03-11 14:11:21 -05:00
#define testBackupPqScriptP(pgVersion, backupStartTime, ...) \
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
testBackupPqScript(pgVersion, backupStartTime, (TestBackupPqScriptParam){VAR_PARAM_INIT, __VA_ARGS__})
static void
testBackupPqScript(unsigned int pgVersion, time_t backupTimeStart, TestBackupPqScriptParam param)
{
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
const char *pg1Path = TEST_PATH "/pg1";
const char *pg2Path = TEST_PATH "/pg2";
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Fix error in timeline conversion. The timeline is required to verify WAL segments in the archive after a backup. The conversion was performed base 10 instead of 16, which led to errors when the timeline was ≥ 0xA.
2020-01-21 10:29:46 -07:00
// If no timeline specified then use timeline 1
param.timeline = param.timeline == 0 ? 1 : param.timeline;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Read pg_control to get info about the cluster
PgControl pgControl = pgControlFromFile(storagePg());
// Set archive timeout really small to save time on errors
Replace double type with time in config module. The C code does not use doubles to represent seconds like the Perl code did so time can be represented as an integer which reduces the number of data types that config has to understand. Also remove Variant doubles since they are no longer used. Note that not all double code was removed since we still need to display times to the user in seconds and it is possible for the times to be fractional. In the future this will likely be simplified by storing the original user input and using that value when the time needs to be displayed.
2020-12-09 08:59:51 -05:00
cfgOptionSet(cfgOptArchiveTimeout, cfgSourceParam, varNewInt64(100));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Add timeline and checkpoint checks to backup. Add the following checks: * Checkpoint is updated in pg_control after pg_start_backup(). This helps ensure that PostgreSQL and pgBackRest have a consistent view of the storage and that PGDATA paths match. * Timeline of backup start WAL file matches pg_control. Hard to see how this one could get hit, but we have the power... * Standby is on the same timeline as the primary. If not, this standby is not following the primary. * Last standby checkpoint is not greater than the backup checkpoint. If so, this standby is not following the primary. This also requires some additional plumbing to read/write timeline/checkpoint from pg_control and parse timelines from WAL filenames. There were some changes in the backup tests caused by the fact that pg_control now has different contents for each backup. The check to ensure that the required checkpoint was reached on the standby should also be updated to use pg_control (it currently uses pg_control_checkpoint()), but that requires non-trivial changes to the test harness and will need to wait.
2021-12-07 09:21:07 -05:00
// Set LSN and WAL start/stop
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
uint64_t lsnStart = ((uint64_t)backupTimeStart & 0xFFFFFF00) << 28;
uint64_t lsnStop =
lsnStart + ((param.walTotal == 0 ? 0 : param.walTotal - 1) * pgControl.walSegmentSize) + (pgControl.walSegmentSize / 2);
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
const char *walSegmentPrior = strZ(
pgLsnToWalSegment(param.timeline, lsnStart - pgControl.walSegmentSize, pgControl.walSegmentSize));
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
const char *lsnStartStr = strZ(pgLsnToStr(lsnStart));
const char *walSegmentStart = strZ(pgLsnToWalSegment(param.timeline, lsnStart, pgControl.walSegmentSize));
const char *lsnStopStr = strZ(pgLsnToStr(lsnStop));
const char *walSegmentStop = strZ(pgLsnToWalSegment(param.timeline, lsnStop, pgControl.walSegmentSize));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Add timeline and checkpoint checks to backup. Add the following checks: * Checkpoint is updated in pg_control after pg_start_backup(). This helps ensure that PostgreSQL and pgBackRest have a consistent view of the storage and that PGDATA paths match. * Timeline of backup start WAL file matches pg_control. Hard to see how this one could get hit, but we have the power... * Standby is on the same timeline as the primary. If not, this standby is not following the primary. * Last standby checkpoint is not greater than the backup checkpoint. If so, this standby is not following the primary. This also requires some additional plumbing to read/write timeline/checkpoint from pg_control and parse timelines from WAL filenames. There were some changes in the backup tests caused by the fact that pg_control now has different contents for each backup. The check to ensure that the required checkpoint was reached on the standby should also be updated to use pg_control (it currently uses pg_control_checkpoint()), but that requires non-trivial changes to the test harness and will need to wait.
2021-12-07 09:21:07 -05:00
// Save pg_control with updated info
pgControl.checkpoint = lsnStart;
pgControl.timeline = param.timeline;
HRN_STORAGE_PUT(
storagePgIdxWrite(0), PG_PATH_GLOBAL "/" PG_FILE_PGCONTROL, hrnPgControlToBuffer(pgControl),
.timeModified = backupTimeStart);
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
// Update pg_control on primary with the backup time
HRN_PG_CONTROL_TIME(storagePgIdxWrite(0), backupTimeStart);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Write WAL segments to the archive
// -----------------------------------------------------------------------------------------------------------------------------
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
if (!param.noPriorWal)
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
{
InfoArchive *infoArchive = infoArchiveLoadFile(storageRepo(), INFO_ARCHIVE_PATH_FILE_STR, cipherTypeNone, NULL);
const String *archiveId = infoArchiveId(infoArchive);
Fix error in timeline conversion. The timeline is required to verify WAL segments in the archive after a backup. The conversion was performed base 10 instead of 16, which led to errors when the timeline was ≥ 0xA.
2020-01-21 10:29:46 -07:00
StringList *walSegmentList = pgLsnRangeToWalSegmentList(
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
pgControl.version, param.timeline, lsnStart - pgControl.walSegmentSize,
param.noWal ? lsnStart - pgControl.walSegmentSize : lsnStop,
pgControl.walSegmentSize);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Buffer *walBuffer = bufNew((size_t)pgControl.walSegmentSize);
bufUsedSet(walBuffer, bufSize(walBuffer));
memset(bufPtr(walBuffer), 0, bufSize(walBuffer));
Move PostgreSQL version interface test functions to a test harness. Some version interface test functions were integrated into the core code because they relied on the PostgreSQL versioned interface. Even though they were compiled out for production builds they cluttered the core code and made it harder to determine what was required by core. Create a PostgreSQL version interface in a test harness to contain these functions. This does require some duplication but the cleaner core code seems a good tradeoff. It is possible for some of this code to be auto-generated but since it is only updated once per year the matter is not pressing.
2021-05-17 07:20:28 -04:00
hrnPgWalToBuffer((PgWal){.version = pgControl.version, .systemId = pgControl.systemId}, walBuffer);
Use StringId for type in cryptoHash() and cryptoHmacOne(). This brings hash types in line with cipher types (i.e. a StringId enum) and allows removal of some Strings.
2022-05-25 15:27:53 -04:00
const String *walChecksum = bufHex(cryptoHashOne(hashTypeSha1, walBuffer));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
for (unsigned int walSegmentIdx = 0; walSegmentIdx < strLstSize(walSegmentList); walSegmentIdx++)
{
StorageWrite *write = storageNewWriteP(
storageRepoWrite(),
strNewFmt(
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
STORAGE_REPO_ARCHIVE "/%s/%s-%s%s", strZ(archiveId), strZ(strLstGet(walSegmentList, walSegmentIdx)),
strZ(walChecksum), strZ(compressExtStr(param.walCompressType))));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Add infrastructure for multiple compression type support. Add compress-type option and deprecate compress option. Since the compress option is boolean it won't work with multiple compression types. Add logic to cfgLoadUpdateOption() to update compress-type if it is not set directly. The compress option should no longer be referenced outside the cfgLoadUpdateOption() function. Add common/compress/helper module to contain interface functions that work with multiple compression types. Code outside this module should no longer call specific compression drivers, though it may be OK to reference a specific compression type using the new interface (e.g., saving backup history files in gz format). Unit tests only test compression using the gz format because other formats may not be available in all builds. It is the job of integration tests to exercise all compression types. Additional compression types will be added in future commits.
2020-03-06 14:41:03 -05:00
if (param.walCompressType != compressTypeNone)
ioFilterGroupAdd(ioWriteFilterGroup(storageWriteIo(write)), compressFilter(param.walCompressType, 1));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
storagePutP(write, walBuffer);
}
}
// -----------------------------------------------------------------------------------------------------------------------------
if (pgVersion == PG_VERSION_95)
{
ASSERT(!param.backupStandby);
ASSERT(!param.errorAfterStart);
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
if (param.noArchiveCheck)
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
{
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
harnessPqScriptSet((HarnessPq [])
{
// Connect to primary
HRNPQ_MACRO_OPEN_GE_92(1, "dbname='postgres' port=5432", PG_VERSION_95, pg1Path, false, NULL, NULL),
// Get start time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Start backup
HRNPQ_MACRO_ADVISORY_LOCK(1, true),
HRNPQ_MACRO_IS_IN_BACKUP(1, false),
Remove support for PostgreSQL 8.3/8.4. There is no evidence that users need 8.3/8.4 anymore but it does cost us in terms of development and testing, especially now that we have a number of new backup/restore features planned. It seems to make sense to remove this support now. If there are users who need to use/migrate from these versions they can use an older version of pgBackRest.
2022-01-06 15:34:04 -05:00
HRNPQ_MACRO_START_BACKUP_LE_95(1, param.startFast, lsnStartStr, walSegmentStart),
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
HRNPQ_MACRO_DATABASE_LIST_1(1, "test1"),
HRNPQ_MACRO_TABLESPACE_LIST_0(1),
// Get copy start time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 999),
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 1000),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check that clusters are alive and correctly configured during a backup. Fail the backup if a cluster stops or the standby is promoted. Previously, shutting down the primary would cause an error but it was not detected until the end of the backup. Now the error will happen sooner and a promotion on the standby will also cause an error.
2021-12-08 10:16:41 -05:00
// Ping
HRNPQ_MACRO_IS_STANDBY_QUERY(1, false),
HRNPQ_MACRO_IS_STANDBY_QUERY(1, false),
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Stop backup
HRNPQ_MACRO_STOP_BACKUP_LE_95(1, lsnStopStr, walSegmentStop),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Get stop time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 2000),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
HRNPQ_MACRO_DONE()
});
}
else
{
harnessPqScriptSet((HarnessPq [])
{
// Connect to primary
HRNPQ_MACRO_OPEN_GE_92(1, "dbname='postgres' port=5432", PG_VERSION_95, pg1Path, false, NULL, NULL),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Get start time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Start backup
HRNPQ_MACRO_ADVISORY_LOCK(1, true),
HRNPQ_MACRO_IS_IN_BACKUP(1, false),
HRNPQ_MACRO_CURRENT_WAL_LE_96(1, walSegmentPrior),
Remove support for PostgreSQL 8.3/8.4. There is no evidence that users need 8.3/8.4 anymore but it does cost us in terms of development and testing, especially now that we have a number of new backup/restore features planned. It seems to make sense to remove this support now. If there are users who need to use/migrate from these versions they can use an older version of pgBackRest.
2022-01-06 15:34:04 -05:00
HRNPQ_MACRO_START_BACKUP_LE_95(1, param.startFast, lsnStartStr, walSegmentStart),
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
HRNPQ_MACRO_DATABASE_LIST_1(1, "test1"),
HRNPQ_MACRO_TABLESPACE_LIST_0(1),
// Get copy start time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 999),
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 1000),
Check that clusters are alive and correctly configured during a backup. Fail the backup if a cluster stops or the standby is promoted. Previously, shutting down the primary would cause an error but it was not detected until the end of the backup. Now the error will happen sooner and a promotion on the standby will also cause an error.
2021-12-08 10:16:41 -05:00
// Ping
HRNPQ_MACRO_IS_STANDBY_QUERY(1, false),
HRNPQ_MACRO_IS_STANDBY_QUERY(1, false),
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Stop backup
HRNPQ_MACRO_STOP_BACKUP_LE_95(1, lsnStopStr, walSegmentStop),
// Get stop time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 2000),
HRNPQ_MACRO_DONE()
});
}
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// -----------------------------------------------------------------------------------------------------------------------------
else if (pgVersion == PG_VERSION_96)
{
ASSERT(param.backupStandby);
ASSERT(!param.errorAfterStart);
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
ASSERT(!param.noArchiveCheck);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Add timeline and checkpoint checks to backup. Add the following checks: * Checkpoint is updated in pg_control after pg_start_backup(). This helps ensure that PostgreSQL and pgBackRest have a consistent view of the storage and that PGDATA paths match. * Timeline of backup start WAL file matches pg_control. Hard to see how this one could get hit, but we have the power... * Standby is on the same timeline as the primary. If not, this standby is not following the primary. * Last standby checkpoint is not greater than the backup checkpoint. If so, this standby is not following the primary. This also requires some additional plumbing to read/write timeline/checkpoint from pg_control and parse timelines from WAL filenames. There were some changes in the backup tests caused by the fact that pg_control now has different contents for each backup. The check to ensure that the required checkpoint was reached on the standby should also be updated to use pg_control (it currently uses pg_control_checkpoint()), but that requires non-trivial changes to the test harness and will need to wait.
2021-12-07 09:21:07 -05:00
// Save pg_control with updated info
HRN_STORAGE_PUT(storagePgIdxWrite(1), PG_PATH_GLOBAL "/" PG_FILE_PGCONTROL, hrnPgControlToBuffer(pgControl));
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
if (param.noPriorWal)
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
{
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
harnessPqScriptSet((HarnessPq [])
{
// Connect to primary
HRNPQ_MACRO_OPEN_GE_96(1, "dbname='postgres' port=5432", PG_VERSION_96, pg1Path, false, NULL, NULL),
// Connect to standby
HRNPQ_MACRO_OPEN_GE_96(2, "dbname='postgres' port=5433", PG_VERSION_96, pg2Path, true, NULL, NULL),
// Get start time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000),
// Start backup
HRNPQ_MACRO_ADVISORY_LOCK(1, true),
HRNPQ_MACRO_CURRENT_WAL_LE_96(1, walSegmentPrior),
HRNPQ_MACRO_START_BACKUP_96(1, true, lsnStartStr, walSegmentStart),
HRNPQ_MACRO_DATABASE_LIST_1(1, "test1"),
HRNPQ_MACRO_TABLESPACE_LIST_0(1),
// Wait for standby to sync
HRNPQ_MACRO_REPLAY_WAIT_96(2, lsnStartStr),
HRNPQ_MACRO_DONE()
});
}
else
{
harnessPqScriptSet((HarnessPq [])
{
// Connect to primary
HRNPQ_MACRO_OPEN_GE_96(1, "dbname='postgres' port=5432", PG_VERSION_96, pg1Path, false, NULL, NULL),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Connect to standby
HRNPQ_MACRO_OPEN_GE_96(2, "dbname='postgres' port=5433", PG_VERSION_96, pg2Path, true, NULL, NULL),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Get start time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Start backup
HRNPQ_MACRO_ADVISORY_LOCK(1, true),
HRNPQ_MACRO_CURRENT_WAL_LE_96(1, walSegmentPrior),
HRNPQ_MACRO_START_BACKUP_96(1, true, lsnStartStr, walSegmentStart),
HRNPQ_MACRO_DATABASE_LIST_1(1, "test1"),
HRNPQ_MACRO_TABLESPACE_LIST_0(1),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Wait for standby to sync
HRNPQ_MACRO_REPLAY_WAIT_96(2, lsnStartStr),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Get copy start time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 999),
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 1000),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check that clusters are alive and correctly configured during a backup. Fail the backup if a cluster stops or the standby is promoted. Previously, shutting down the primary would cause an error but it was not detected until the end of the backup. Now the error will happen sooner and a promotion on the standby will also cause an error.
2021-12-08 10:16:41 -05:00
// Ping
HRNPQ_MACRO_IS_STANDBY_QUERY(1, false),
HRNPQ_MACRO_IS_STANDBY_QUERY(2, true),
HRNPQ_MACRO_IS_STANDBY_QUERY(1, false),
HRNPQ_MACRO_IS_STANDBY_QUERY(2, true),
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Stop backup
HRNPQ_MACRO_STOP_BACKUP_96(1, lsnStopStr, walSegmentStop, false),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Get stop time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 2000),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
HRNPQ_MACRO_DONE()
});
}
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// -----------------------------------------------------------------------------------------------------------------------------
else if (pgVersion == PG_VERSION_11)
{
ASSERT(!param.backupStandby);
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
ASSERT(!param.noArchiveCheck);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
if (param.errorAfterStart)
{
harnessPqScriptSet((HarnessPq [])
{
// Connect to primary
Disable query parallelism in PostgreSQL sessions used for backup control. There is no need to have parallelism enabled in a backup control session. In particular, 9.6 marks pg_stop_backup() as parallel-safe but an error will be thrown if pg_stop_backup() is run in a worker.
2020-06-25 08:02:48 -04:00
HRNPQ_MACRO_OPEN_GE_96(1, "dbname='postgres' port=5432", PG_VERSION_11, pg1Path, false, NULL, NULL),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Get start time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000),
// Start backup
HRNPQ_MACRO_ADVISORY_LOCK(1, true),
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
HRNPQ_MACRO_CURRENT_WAL_GE_10(1, walSegmentPrior),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
HRNPQ_MACRO_START_BACKUP_GE_10(1, param.startFast, lsnStartStr, walSegmentStart),
HRNPQ_MACRO_DATABASE_LIST_1(1, "test1"),
HRNPQ_MACRO_TABLESPACE_LIST_1(1, 32768, "tblspc32768"),
// Get copy start time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 999),
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 1000),
HRNPQ_MACRO_DONE()
});
}
else
{
harnessPqScriptSet((HarnessPq [])
{
// Connect to primary
Disable query parallelism in PostgreSQL sessions used for backup control. There is no need to have parallelism enabled in a backup control session. In particular, 9.6 marks pg_stop_backup() as parallel-safe but an error will be thrown if pg_stop_backup() is run in a worker.
2020-06-25 08:02:48 -04:00
HRNPQ_MACRO_OPEN_GE_96(1, "dbname='postgres' port=5432", PG_VERSION_11, pg1Path, false, NULL, NULL),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Get start time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000),
// Start backup
HRNPQ_MACRO_ADVISORY_LOCK(1, true),
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
HRNPQ_MACRO_CURRENT_WAL_GE_10(1, walSegmentStart),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
HRNPQ_MACRO_START_BACKUP_GE_10(1, param.startFast, lsnStartStr, walSegmentStart),
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Switch WAL segment so it can be checked
HRNPQ_MACRO_CREATE_RESTORE_POINT(1, "X/X"),
HRNPQ_MACRO_WAL_SWITCH(1, "wal", walSegmentStart),
// Get database and tablespace list
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRNPQ_MACRO_DATABASE_LIST_1(1, "test1"),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
HRNPQ_MACRO_TABLESPACE_LIST_1(1, 32768, "tblspc32768"),
// Get copy start time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 999),
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 1000),
Check that clusters are alive and correctly configured during a backup. Fail the backup if a cluster stops or the standby is promoted. Previously, shutting down the primary would cause an error but it was not detected until the end of the backup. Now the error will happen sooner and a promotion on the standby will also cause an error.
2021-12-08 10:16:41 -05:00
// Ping
HRNPQ_MACRO_IS_STANDBY_QUERY(1, false),
HRNPQ_MACRO_IS_STANDBY_QUERY(1, false),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Stop backup
Add tablespace_map file to command/backup test module. The code worked fine but better to have explicit tests for this file.
2021-11-15 14:32:22 -05:00
HRNPQ_MACRO_STOP_BACKUP_GE_10(1, lsnStopStr, walSegmentStop, true),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Get stop time
HRNPQ_MACRO_TIME_QUERY(1, (int64_t)backupTimeStart * 1000 + 2000),
HRNPQ_MACRO_DONE()
});
}
}
else
THROW_FMT(AssertError, "unsupported test version %u", pgVersion); // {uncoverable - no invalid versions in tests}
};
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
/***********************************************************************************************************************************
Wrap cmdBackup() with lock acquire and release
***********************************************************************************************************************************/
void testCmdBackup(void)
{
FUNCTION_HARNESS_VOID();
lockAcquire(TEST_PATH_STR, cfgOptionStr(cfgOptStanza), cfgOptionStr(cfgOptExecId), lockTypeBackup, 0, true);
TRY_BEGIN()
{
cmdBackup();
}
FINALLY()
{
lockRelease(true);
}
TRY_END();
FUNCTION_HARNESS_RETURN_VOID();
}
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
/***********************************************************************************************************************************
Test Run
***********************************************************************************************************************************/
Add missing static keywords in test modules.
2021-10-18 12:22:48 -04:00
static void
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
testRun(void)
{
FUNCTION_HARNESS_VOID();
Add local process shim. Run the local process inside a forked child process instead of exec'ing it. This allows coverage to accumulate in the local process rather than needing to test the local protocol functions directly, resulting in better end-to-end testing and less test duplication. Another advantage is that the pgbackrest binary does not need to be built for the test. The backup, restore, and verify command tests have been updated to use the new shim for coverage.
2021-05-21 12:45:00 -04:00
// Install local command handler shim
static const ProtocolServerHandler testLocalHandlerList[] = {PROTOCOL_SERVER_HANDLER_BACKUP_LIST};
Add LENGTH_OF() macro. Determining the length of arrays that could be calculated at compile time was a bit piecemeal, with special macros used sometimes and with the math done directly other times. This macro makes the task easier, uses less space, and automatically adjusts when the type changes.
2022-04-07 19:00:15 -04:00
hrnProtocolLocalShimInstall(testLocalHandlerList, LENGTH_OF(testLocalHandlerList));
Add local process shim. Run the local process inside a forked child process instead of exec'ing it. This allows coverage to accumulate in the local process rather than needing to test the local protocol functions directly, resulting in better end-to-end testing and less test duplication. Another advantage is that the pgbackrest binary does not need to be built for the test. The backup, restore, and verify command tests have been updated to use the new shim for coverage.
2021-05-21 12:45:00 -04:00
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// The tests expect the timezone to be UTC
setenv("TZ", "UTC", true);
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
Storage *storageTest = storagePosixNewP(TEST_PATH_STR, .write = true);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
// *****************************************************************************************************************************
if (testBegin("backupRegExp()"))
{
const String *full = STRDEF("20181119-152138F");
const String *incr = STRDEF("20181119-152138F_20181119-152152I");
const String *diff = STRDEF("20181119-152138F_20181119-152152D");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("regular expression - error");
TEST_ERROR(
backupRegExpP(0),
AssertError, "assertion 'param.full || param.differential || param.incremental' failed");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("regular expression - match full");
String *filter = backupRegExpP(.full = true);
TEST_RESULT_STR_Z(filter, "^[0-9]{8}\\-[0-9]{6}F$", "full backup regex with anchors");
TEST_RESULT_BOOL(regExpMatchOne(filter, incr), false, "does not exactly match incr");
TEST_RESULT_BOOL(regExpMatchOne(filter, diff), false, "does not exactly match diff");
TEST_RESULT_BOOL(regExpMatchOne(filter, full), true, "exactly matches full");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("regular expression - match full, incremental");
filter = backupRegExpP(.full = true, .incremental = true);
TEST_RESULT_STR_Z(
filter, "^[0-9]{8}\\-[0-9]{6}F(\\_[0-9]{8}\\-[0-9]{6}I){0,1}$", "full and optional incr backup regex with anchors");
TEST_RESULT_BOOL(regExpMatchOne(filter, incr), true, "match incr");
TEST_RESULT_BOOL(regExpMatchOne(filter, diff), false, "does not match diff");
TEST_RESULT_BOOL(regExpMatchOne(filter, full), true, "match full");
TEST_RESULT_BOOL(
regExpMatchOne(
filter, STRDEF("12341234-123123F_12341234-123123IG")), false, "does not match with trailing character");
TEST_RESULT_BOOL(
regExpMatchOne(
filter, STRDEF("A12341234-123123F_12341234-123123I")), false, "does not match with leading character");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("regular expression - match full, differential");
filter = backupRegExpP(.full = true, .differential = true);
TEST_RESULT_STR_Z(
filter, "^[0-9]{8}\\-[0-9]{6}F(\\_[0-9]{8}\\-[0-9]{6}D){0,1}$", "full and optional diff backup regex with anchors");
TEST_RESULT_BOOL(regExpMatchOne(filter, incr), false, "does not match incr");
TEST_RESULT_BOOL(regExpMatchOne(filter, diff), true, "match diff");
TEST_RESULT_BOOL(regExpMatchOne(filter, full), true, "match full");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("regular expression - match full, incremental, differential");
filter = backupRegExpP(.full = true, .incremental = true, .differential = true);
TEST_RESULT_STR_Z(
filter, "^[0-9]{8}\\-[0-9]{6}F(\\_[0-9]{8}\\-[0-9]{6}(D|I)){0,1}$",
"full, optional diff and incr backup regex with anchors");
TEST_RESULT_BOOL(regExpMatchOne(filter, incr), true, "match incr");
TEST_RESULT_BOOL(regExpMatchOne(filter, diff), true, "match diff");
TEST_RESULT_BOOL(regExpMatchOne(filter, full), true, "match full");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("regular expression - match incremental, differential without end anchor");
filter = backupRegExpP(.incremental = true, .differential = true, .noAnchorEnd = true);
TEST_RESULT_STR_Z(filter, "^[0-9]{8}\\-[0-9]{6}F\\_[0-9]{8}\\-[0-9]{6}(D|I)", "diff and incr backup regex with anchors");
TEST_RESULT_BOOL(regExpMatchOne(filter, incr), true, "match incr");
TEST_RESULT_BOOL(regExpMatchOne(filter, diff), true, "match diff");
TEST_RESULT_BOOL(regExpMatchOne(filter, full), false, "does not match full");
TEST_RESULT_BOOL(
regExpMatchOne(
filter, STRDEF("A12341234-123123F_12341234-123123I")), false, "does not match with leading character");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("regular expression - match incremental");
filter = backupRegExpP(.incremental = true);
TEST_RESULT_STR_Z(filter, "^[0-9]{8}\\-[0-9]{6}F\\_[0-9]{8}\\-[0-9]{6}I$", "incr backup regex with anchors");
TEST_RESULT_BOOL(regExpMatchOne(filter, incr), true, "match incr");
TEST_RESULT_BOOL(regExpMatchOne(filter, diff), false, "does not match diff");
TEST_RESULT_BOOL(regExpMatchOne(filter, full), false, "does not match full");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("regular expression - match differential");
filter = backupRegExpP(.differential = true);
TEST_RESULT_STR_Z(filter, "^[0-9]{8}\\-[0-9]{6}F\\_[0-9]{8}\\-[0-9]{6}D$", "diff backup regex with anchors");
TEST_RESULT_BOOL(regExpMatchOne(filter, incr), false, "does not match incr");
TEST_RESULT_BOOL(regExpMatchOne(filter, diff), true, "match diff");
TEST_RESULT_BOOL(regExpMatchOne(filter, full), false, "does not match full");
}
// *****************************************************************************************************************************
if (testBegin("PageChecksum"))
{
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("segment page default");
TEST_RESULT_UINT(PG_SEGMENT_PAGE_DEFAULT, 131072, "check pages per segment");
// -------------------------------------------------------------------------------------------------------------------------
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
TEST_TITLE("two misaligned buffers in a row");
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
Buffer *buffer = bufNew(513);
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
bufUsedSet(buffer, bufSize(buffer));
memset(bufPtr(buffer), 0, bufSize(buffer));
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
*(PageHeaderData *)(bufPtr(buffer) + (PG_PAGE_SIZE_DEFAULT * 0x00)) = (PageHeaderData){.pd_upper = 0};
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
Buffer *bufferOut = bufNew(513);
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
IoWrite *write = ioBufferWriteNew(bufferOut);
ioFilterGroupAdd(
ioWriteFilterGroup(write),
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
pageChecksumNewPack(ioFilterParamList(pageChecksumNew(0, PG_SEGMENT_PAGE_DEFAULT, STRDEF(BOGUS_STR)))));
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
ioWriteOpen(write);
ioWrite(write, buffer);
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
TEST_ERROR(ioWrite(write, buffer), AssertError, "should not be possible to see two misaligned pages in a row");
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
// -------------------------------------------------------------------------------------------------------------------------
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
TEST_TITLE("retry a page with an invalid checksum");
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
// Write to file with valid checksums
buffer = bufNew(PG_PAGE_SIZE_DEFAULT * 4);
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
memset(bufPtr(buffer), 0, bufSize(buffer));
bufUsedSet(buffer, bufSize(buffer));
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
*(PageHeaderData *)(bufPtr(buffer) + (PG_PAGE_SIZE_DEFAULT * 0x00)) = (PageHeaderData){.pd_upper = 0x00};
*(PageHeaderData *)(bufPtr(buffer) + (PG_PAGE_SIZE_DEFAULT * 0x01)) = (PageHeaderData){.pd_upper = 0xFF};
((PageHeaderData *)(bufPtr(buffer) + (PG_PAGE_SIZE_DEFAULT * 0x01)))->pd_checksum = pgPageChecksum(
bufPtr(buffer) + (PG_PAGE_SIZE_DEFAULT * 0x01), 1);
*(PageHeaderData *)(bufPtr(buffer) + (PG_PAGE_SIZE_DEFAULT * 0x02)) = (PageHeaderData){.pd_upper = 0x00};
*(PageHeaderData *)(bufPtr(buffer) + (PG_PAGE_SIZE_DEFAULT * 0x03)) = (PageHeaderData){.pd_upper = 0xFE};
((PageHeaderData *)(bufPtr(buffer) + (PG_PAGE_SIZE_DEFAULT * 0x03)))->pd_checksum = pgPageChecksum(
bufPtr(buffer) + (PG_PAGE_SIZE_DEFAULT * 0x03), 3);
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
HRN_STORAGE_PUT(storageTest, "relation", buffer);
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
// Now break the checksum to force a retry
((PageHeaderData *)(bufPtr(buffer) + (PG_PAGE_SIZE_DEFAULT * 0x01)))->pd_checksum = 0;
((PageHeaderData *)(bufPtr(buffer) + (PG_PAGE_SIZE_DEFAULT * 0x03)))->pd_checksum = 0;
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
write = ioBufferWriteNew(bufferOut);
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
ioFilterGroupAdd(
ioWriteFilterGroup(write), pageChecksumNew(0, PG_SEGMENT_PAGE_DEFAULT, storagePathP(storageTest, STRDEF("relation"))));
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
ioWriteOpen(write);
ioWrite(write, buffer);
ioWriteClose(write);
TEST_RESULT_STR_Z(
hrnPackToStr(ioFilterGroupResultPackP(ioWriteFilterGroup(write), PAGE_CHECKSUM_FILTER_TYPE)),
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
"2:bool:true, 3:bool:true", "valid on retry");
Move command/backup-common tests in the command/backup module. As much as possible it is better to get coverage with more realistic tests. Merging these modules will allow the page checksum code to be covered with real backups.
2022-02-18 17:50:05 -06:00
}
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// *****************************************************************************************************************************
if (testBegin("segmentNumber()"))
{
Combine encrypted backupFile() tests with unencrypted tests. This makes it easier to comment out all the tests while developing without getting unused variable errors.
2022-01-09 10:11:00 -05:00
TEST_RESULT_UINT(segmentNumber(STRDEF("999")), 0, "No segment number");
TEST_RESULT_UINT(segmentNumber(STRDEF("999.123")), 123, "Segment number");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
}
// *****************************************************************************************************************************
Add local process shim. Run the local process inside a forked child process instead of exec'ing it. This allows coverage to accumulate in the local process rather than needing to test the local protocol functions directly, resulting in better end-to-end testing and less test duplication. Another advantage is that the pgbackrest binary does not need to be built for the test. The backup, restore, and verify command tests have been updated to use the new shim for coverage.
2021-05-21 12:45:00 -04:00
if (testBegin("backupFile()"))
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
{
Combine encrypted backupFile() tests with unencrypted tests. This makes it easier to comment out all the tests while developing without getting unused variable errors.
2022-01-09 10:11:00 -05:00
const String *pgFile = STRDEF("testfile");
const String *missingFile = STRDEF("missing");
const String *backupLabel = STRDEF("20190718-155825F");
const String *backupPathFile = strNewFmt(STORAGE_REPO_BACKUP "/%s/%s", strZ(backupLabel), strZ(pgFile));
BackupFileResult result = {0};
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// Load Parameters
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg");
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// Create the pg path
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PATH_CREATE(storagePgWrite(), NULL, .mode = 0700);
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("pg file missing - ignoreMissing=true");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
List *fileList = lstNewP(sizeof(BackupFile));
BackupFile file =
{
.pgFile = missingFile,
.pgFileIgnoreMissing = true,
.pgFileSize = 0,
.pgFileCopyExactSize = true,
.pgFileChecksum = NULL,
.pgFileChecksumPage = false,
.manifestFile = missingFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
const String *repoFile = strNewFmt(STORAGE_REPO_BACKUP "/%s/%s", strZ(backupLabel), strZ(file.manifestFile));
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeNone, 1, false, cipherTypeNone, NULL, fileList), 0),
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
"pg file missing, ignoreMissing=true, no delta");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.copySize + result.repoSize, 0, "copy/repo size 0");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultSkip, "skip file");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("pg file missing - ignoreMissing=false");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = missingFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 0,
.pgFileCopyExactSize = true,
.pgFileChecksum = NULL,
.pgFileChecksumPage = false,
.manifestFile = missingFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
TEST_ERROR(
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
backupFile(repoFile, compressTypeNone, 1, false, cipherTypeNone, NULL, fileList), FileMissingError,
"unable to open missing file '" TEST_PATH "/pg/missing' for read");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// Create a pg file to backup
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), strZ(pgFile), "atestfile");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
// Remove repo file
HRN_STORAGE_REMOVE(storageRepoWrite(), strZ(backupPathFile));
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("test pagechecksum while db file grows");
Limit backup file copy size to size reported at backup start. If a file grows during the backup it will be reconstructed by WAL replay during recovery so there is no need to copy the additional data. This also reduces the likelihood of seeing torn pages during the copy. Torn pages can still occur in the middle of the file, though, so they must be handled.
2020-03-19 13:16:05 -04:00
// Increase the file size but most of the following tests will still treat the file as size 9. This tests the common case
// where a file grows while a backup is running.
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), strZ(pgFile), "atestfile###");
Limit backup file copy size to size reported at backup start. If a file grows during the backup it will be reconstructed by WAL replay during recovery so there is no need to copy the additional data. This also reduces the likelihood of seeing torn pages during the copy. Torn pages can still occur in the middle of the file, though, so they must be handled.
2020-03-19 13:16:05 -04:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 9,
.pgFileCopyExactSize = true,
.pgFileChecksum = NULL,
.pgFileChecksumPage = true,
.manifestFile = pgFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
repoFile = strNewFmt(STORAGE_REPO_BACKUP "/%s/%s", strZ(backupLabel), strZ(file.manifestFile));
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeNone, 1, false, cipherTypeNone, NULL, fileList), 0),
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
"file checksummed with pageChecksum enabled");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.copySize, 9, "copy=pgFile size");
TEST_RESULT_UINT(result.repoSize, 9, "repo=pgFile size");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultCopy, "copy file");
TEST_RESULT_STR_Z(result.copyChecksum, "9bc8ab2dda60ef4beed07d1e19ce0676d5edde67", "copy checksum matches");
Convert page checksum filter result to a pack. The pack is both more compact and more efficient than a variant. Also aggregate the page error info in the main process rather than in the filter to allow additional LSN filtering, to be added in a future commit.
2021-09-24 17:40:31 -04:00
TEST_RESULT_STR_Z(hrnPackToStr(result.pageChecksumResult), "2:bool:false, 3:bool:false", "pageChecksumResult");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_STORAGE_EXISTS(storageRepoWrite(), strZ(backupPathFile), .remove = true, .comment = "check exists in repo, remove");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
TEST_TITLE("pgFileSize, ignoreMissing=false, backupLabel, pgFileChecksumPage");
Add local process shim. Run the local process inside a forked child process instead of exec'ing it. This allows coverage to accumulate in the local process rather than needing to test the local protocol functions directly, resulting in better end-to-end testing and less test duplication. Another advantage is that the pgbackrest binary does not need to be built for the test. The backup, restore, and verify command tests have been updated to use the new shim for coverage.
2021-05-21 12:45:00 -04:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 8,
.pgFileCopyExactSize = false,
.pgFileChecksum = NULL,
.pgFileChecksumPage = true,
.manifestFile = pgFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
Add local process shim. Run the local process inside a forked child process instead of exec'ing it. This allows coverage to accumulate in the local process rather than needing to test the local protocol functions directly, resulting in better end-to-end testing and less test duplication. Another advantage is that the pgbackrest binary does not need to be built for the test. The backup, restore, and verify command tests have been updated to use the new shim for coverage.
2021-05-21 12:45:00 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeNone, 1, false, cipherTypeNone, NULL, fileList), 0),
Add local process shim. Run the local process inside a forked child process instead of exec'ing it. This allows coverage to accumulate in the local process rather than needing to test the local protocol functions directly, resulting in better end-to-end testing and less test duplication. Another advantage is that the pgbackrest binary does not need to be built for the test. The backup, restore, and verify command tests have been updated to use the new shim for coverage.
2021-05-21 12:45:00 -04:00
"backup file");
TEST_RESULT_UINT(result.copySize, 12, "copy size");
TEST_RESULT_UINT(result.repoSize, 12, "repo size");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultCopy, "copy file");
TEST_RESULT_STR_Z(result.copyChecksum, "c3ae4687ea8ccd47bfdb190dbe7fd3b37545fdb9", "checksum");
Convert page checksum filter result to a pack. The pack is both more compact and more efficient than a variant. Also aggregate the page error info in the main process rather than in the filter to allow additional LSN filtering, to be added in a future commit.
2021-09-24 17:40:31 -04:00
TEST_RESULT_STR_Z(hrnPackToStr(result.pageChecksumResult), "2:bool:false, 3:bool:false", "page checksum");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_STORAGE_GET(storageRepo(), strZ(backupPathFile), "atestfile###");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("file exists in repo and db, checksum match - NOOP");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 9,
.pgFileCopyExactSize = true,
.pgFileChecksum = STRDEF("9bc8ab2dda60ef4beed07d1e19ce0676d5edde67"),
.pgFileChecksumPage = false,
.manifestFile = pgFile,
.manifestFileHasReference = true,
};
lstAdd(fileList, &file);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
// File exists in repo and db, pg checksum match, delta set, ignoreMissing false, hasReference - NOOP
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeNone, 1, true, cipherTypeNone, NULL, fileList), 0),
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
"file in db and repo, checksum equal, no ignoreMissing, no pageChecksum, delta, hasReference");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.copySize, 9, "copy size set");
TEST_RESULT_UINT(result.repoSize, 0, "repo size not set since already exists in repo");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultNoOp, "noop file");
TEST_RESULT_STR_Z(result.copyChecksum, "9bc8ab2dda60ef4beed07d1e19ce0676d5edde67", "copy checksum matches");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum result is NULL");
TEST_STORAGE_GET(storageRepo(), strZ(backupPathFile), "atestfile###", .comment = "file not modified");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("file exists in repo and db, checksum mismatch - COPY");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 9,
.pgFileCopyExactSize = true,
.pgFileChecksum = STRDEF("1234567890123456789012345678901234567890"),
.pgFileChecksumPage = false,
.manifestFile = pgFile,
.manifestFileHasReference = true,
};
lstAdd(fileList, &file);
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// File exists in repo and db, pg checksum mismatch, delta set, ignoreMissing false, hasReference - COPY
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeNone, 1, true, cipherTypeNone, NULL, fileList), 0),
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
"file in db and repo, pg checksum not equal, no ignoreMissing, no pageChecksum, delta, hasReference");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.copySize, 9, "copy 9 bytes");
TEST_RESULT_UINT(result.repoSize, 9, "repo=copy size");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultCopy, "copy file");
TEST_RESULT_STR_Z(result.copyChecksum, "9bc8ab2dda60ef4beed07d1e19ce0676d5edde67", "copy checksum for file size 9");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum result is NULL");
TEST_STORAGE_GET(storageRepo(), strZ(backupPathFile), "atestfile", .comment = "9 bytes copied");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("file exists in repo and pg, copy only exact file even if size passed is greater - COPY");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 9999999,
.pgFileCopyExactSize = true,
.pgFileChecksum = STRDEF("9bc8ab2dda60ef4beed07d1e19ce0676d5edde67"),
.pgFileChecksumPage = false,
.manifestFile = pgFile,
.manifestFileHasReference = true,
};
lstAdd(fileList, &file);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
// File exists in repo and pg, pg checksum same, pg size passed is different, delta set, ignoreMissing false, hasReference
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeNone, 1, true, cipherTypeNone, NULL, fileList), 0),
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
"db & repo file, pg checksum same, pg size different, no ignoreMissing, no pageChecksum, delta, hasReference");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.copySize, 12, "copy=pgFile size");
TEST_RESULT_UINT(result.repoSize, 12, "repo=pgFile size");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultCopy, "copy file");
TEST_RESULT_STR_Z(result.copyChecksum, "c3ae4687ea8ccd47bfdb190dbe7fd3b37545fdb9", "copy checksum updated");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum result is NULL");
TEST_STORAGE_GET(storageRepo(), strZ(backupPathFile), "atestfile###", .comment = "confirm contents");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
Recopy during backup when resumed file is missing or corrupt. A recopy would occur if the size or checksum was invalid but on error the backup would terminate. Instead, recopy the resumed file on any error. If the error is systemic (e.g. network failure) then it should show up again during the recopy.
2019-12-07 09:48:33 -05:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("resumed file is missing in repo but present in resumed manifest, file same name in repo - RECOPY");
Recopy during backup when resumed file is missing or corrupt. A recopy would occur if the size or checksum was invalid but on error the backup would terminate. Instead, recopy the resumed file on any error. If the error is systemic (e.g. network failure) then it should show up again during the recopy.
2019-12-07 09:48:33 -05:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 9,
.pgFileCopyExactSize = true,
.pgFileChecksum = STRDEF("9bc8ab2dda60ef4beed07d1e19ce0676d5edde67"),
.pgFileChecksumPage = false,
.manifestFile = STRDEF(BOGUS_STR),
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
repoFile = strNewFmt(STORAGE_REPO_BACKUP "/%s/%s", strZ(backupLabel), strZ(file.manifestFile));
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_STORAGE_LIST(
storageRepo(), STORAGE_REPO_BACKUP "/20190718-155825F", "testfile\n", .comment = "resumed file is missing in repo");
Recopy during backup when resumed file is missing or corrupt. A recopy would occur if the size or checksum was invalid but on error the backup would terminate. Instead, recopy the resumed file on any error. If the error is systemic (e.g. network failure) then it should show up again during the recopy.
2019-12-07 09:48:33 -05:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeNone, 1, true, cipherTypeNone, NULL, fileList), 0),
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
"backup 9 bytes of pgfile to file to resume in repo");
TEST_RESULT_UINT(result.copySize, 9, "copy 9 bytes");
TEST_RESULT_UINT(result.repoSize, 9, "repo=copy size");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultReCopy, "check recopy result");
TEST_RESULT_STR_Z(result.copyChecksum, "9bc8ab2dda60ef4beed07d1e19ce0676d5edde67", "copy checksum for file size 9");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum result is NULL");
TEST_STORAGE_GET(
storageRepo(), strZ(backupPathFile), "atestfile###", .comment = "existing file with same name as pgFile not modified");
TEST_STORAGE_GET(
storageRepo(), STORAGE_REPO_BACKUP "/20190718-155825F/" BOGUS_STR, "atestfile", .comment = "resumed file copied");
Recopy during backup when resumed file is missing or corrupt. A recopy would occur if the size or checksum was invalid but on error the backup would terminate. Instead, recopy the resumed file on any error. If the error is systemic (e.g. network failure) then it should show up again during the recopy.
2019-12-07 09:48:33 -05:00
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("file exists in repo & db, checksum not same in repo - RECOPY");
HRN_STORAGE_PUT_Z(
storageRepoWrite(), strZ(backupPathFile), "adifferentfile",
.comment = "create different file (size and checksum) with same name in repo");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 9,
.pgFileCopyExactSize = true,
.pgFileChecksum = STRDEF("9bc8ab2dda60ef4beed07d1e19ce0676d5edde67"),
.pgFileChecksumPage = false,
.manifestFile = pgFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
repoFile = strNewFmt(STORAGE_REPO_BACKUP "/%s/%s", strZ(backupLabel), strZ(file.manifestFile));
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
// Delta set, ignoreMissing false, no hasReference
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeNone, 1, true, cipherTypeNone, NULL, fileList), 0),
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
"db & repo file, pgFileMatch, repo checksum no match, no ignoreMissing, no pageChecksum, delta, no hasReference");
TEST_RESULT_UINT(result.copySize, 9, "copy 9 bytes");
TEST_RESULT_UINT(result.repoSize, 9, "repo=copy size");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultReCopy, "recopy file");
TEST_RESULT_STR_Z(result.copyChecksum, "9bc8ab2dda60ef4beed07d1e19ce0676d5edde67", "copy checksum for file size 9");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum result is NULL");
TEST_STORAGE_GET(storageRepo(), strZ(backupPathFile), "atestfile", .comment = "existing file recopied");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("file exists in repo but missing from db, checksum same in repo - SKIP");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = missingFile,
.pgFileIgnoreMissing = true,
.pgFileSize = 9,
.pgFileCopyExactSize = true,
.pgFileChecksum = STRDEF("9bc8ab2dda60ef4beed07d1e19ce0676d5edde67"),
.pgFileChecksumPage = false,
.manifestFile = pgFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeNone, 1, true, cipherTypeNone, NULL, fileList), 0),
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
"file in repo only, checksum in repo equal, ignoreMissing=true, no pageChecksum, delta, no hasReference");
TEST_RESULT_UINT(result.copySize + result.repoSize, 0, "copy=repo=0 size");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultSkip, "skip file");
TEST_RESULT_PTR(result.copyChecksum, NULL, "copy checksum NULL");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum result is NULL");
TEST_STORAGE_LIST(
storageRepo(), STORAGE_REPO_BACKUP "/20190718-155825F", BOGUS_STR "\n", .comment = "file removed from repo");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("compression set, all other boolean parameters false - COPY");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 9,
.pgFileCopyExactSize = true,
.pgFileChecksum = NULL,
.pgFileChecksumPage = false,
.manifestFile = pgFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
repoFile = strNewFmt(STORAGE_REPO_BACKUP "/%s/%s.gz", strZ(backupLabel), strZ(file.manifestFile));
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeGz, 3, false, cipherTypeNone, NULL, fileList), 0),
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
"pg file exists, no checksum, no ignoreMissing, compression, no pageChecksum, no delta, no hasReference");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.copySize, 9, "copy=pgFile size");
TEST_RESULT_UINT(result.repoSize, 29, "repo compress size");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultCopy, "copy file");
TEST_RESULT_STR_Z(result.copyChecksum, "9bc8ab2dda60ef4beed07d1e19ce0676d5edde67", "copy checksum");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum result is NULL");
TEST_STORAGE_EXISTS(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepo(), zNewFmt(STORAGE_REPO_BACKUP "/%s/%s.gz", strZ(backupLabel), strZ(pgFile)),
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
.comment = "copy file to repo compress success");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("pg and repo file exist & match, prior checksum, compression - COPY CHECKSUM");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 9,
.pgFileCopyExactSize = true,
.pgFileChecksum = STRDEF("9bc8ab2dda60ef4beed07d1e19ce0676d5edde67"),
.pgFileChecksumPage = false,
.manifestFile = pgFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeGz, 3, false, cipherTypeNone, NULL, fileList), 0),
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
"pg file & repo exists, match, checksum, no ignoreMissing, compression, no pageChecksum, no delta, no hasReference");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.copySize, 9, "copy=pgFile size");
Remove logic that tried to determine additional file system compression. In theory, the additional stat() call after a file has been copied to the repo can determine if additional compression has been applied by the file system. However, it has been a very long time since we tested this in practice. There are currently no unit tests that accurately test this feature since it requires a compressed file system like ZFS to work, which never seemed worth the extra cost. It can also add a lot of time to backups if there are a large quantity of small files. In addition, it stands as a blocker for combining files for small file support since it is no longer possible to get per-file sizes from the viewpoint of the file system. There are several ways this could be reworked but none of them are easy while at the same time maintaining current info functionality. It doesn't seem worth keeping an untested feature that will only work in some special cases (if it still works) when it is blocking development.
2022-02-09 09:32:23 -06:00
TEST_RESULT_UINT(result.repoSize, 0, "repo size not calculated");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultChecksum, "checksum file");
TEST_RESULT_STR_Z(result.copyChecksum, "9bc8ab2dda60ef4beed07d1e19ce0676d5edde67", "compressed repo file checksum matches");
TEST_STORAGE_EXISTS(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepo(), zNewFmt(STORAGE_REPO_BACKUP "/%s/%s.gz", strZ(backupLabel), strZ(pgFile)),
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
.comment = "compressed file exists");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("create a zero sized file - checksum will be set but in backupManifestUpdate it will not be copied");
// Create zero sized file in pg
HRN_STORAGE_PUT_EMPTY(storagePgWrite(), "zerofile");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = STRDEF("zerofile"),
.pgFileIgnoreMissing = false,
.pgFileSize = 0,
.pgFileCopyExactSize = true,
.pgFileChecksum = NULL,
.pgFileChecksumPage = false,
.manifestFile = STRDEF("zerofile"),
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
repoFile = strNewFmt(STORAGE_REPO_BACKUP "/%s/%s", strZ(backupLabel), strZ(file.manifestFile));
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// No prior checksum, no compression, no pageChecksum, no delta, no hasReference
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(backupFile(repoFile, compressTypeNone, 1, false, cipherTypeNone, NULL, fileList), 0),
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
"zero-sized pg file exists, no repo file, no ignoreMissing, no pageChecksum, no delta, no hasReference");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.copySize + result.repoSize, 0, "copy=repo=pgFile size 0");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultCopy, "copy file");
TEST_RESULT_PTR_NE(result.copyChecksum, NULL, "checksum set");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum result is NULL");
TEST_STORAGE_LIST(
storageRepo(), STORAGE_REPO_BACKUP "/20190718-155825F",
BOGUS_STR "\n"
"testfile.gz\n"
"zerofile\n",
.comment = "copy zero file to repo success");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
Combine encrypted backupFile() tests with unencrypted tests. This makes it easier to comment out all the tests while developing without getting unused variable errors.
2022-01-09 10:11:00 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("copy file to encrypted repo");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// Load Parameters
Combine encrypted backupFile() tests with unencrypted tests. This makes it easier to comment out all the tests while developing without getting unused variable errors.
2022-01-09 10:11:00 -05:00
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg");
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
hrnCfgArgRawStrId(argList, cfgOptRepoCipherType, cipherTypeAes256Cbc);
hrnCfgEnvRawZ(cfgOptRepoCipherPass, TEST_CIPHER_PASS);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgEnvRemoveRaw(cfgOptRepoCipherPass);
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
// Create the pg path and pg file to backup
HRN_STORAGE_PUT_Z(storagePgWrite(), strZ(pgFile), "atestfile");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 9,
.pgFileCopyExactSize = true,
.pgFileChecksum = NULL,
.pgFileChecksumPage = false,
.manifestFile = pgFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
repoFile = strNewFmt(STORAGE_REPO_BACKUP "/%s/%s", strZ(backupLabel), strZ(file.manifestFile));
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// No prior checksum, no compression, no pageChecksum, no delta, no hasReference
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(
backupFile(repoFile, compressTypeNone, 1, false, cipherTypeAes256Cbc, STRDEF(TEST_CIPHER_PASS), fileList), 0),
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
"pg file exists, no repo file, no ignoreMissing, no pageChecksum, no delta, no hasReference");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.copySize, 9, "copy size set");
TEST_RESULT_UINT(result.repoSize, 32, "repo size set");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultCopy, "copy file");
TEST_RESULT_STR_Z(result.copyChecksum, "9bc8ab2dda60ef4beed07d1e19ce0676d5edde67", "copy checksum");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum NULL");
TEST_STORAGE_GET(
storageRepo(), strZ(backupPathFile), "atestfile", .cipherType = cipherTypeAes256Cbc,
.comment = "copy file to encrypted repo success");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("delta, copy file (size mismatch) to encrypted repo");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 8,
.pgFileCopyExactSize = true,
.pgFileChecksum = STRDEF("9bc8ab2dda60ef4beed07d1e19ce0676d5edde67"),
.pgFileChecksumPage = false,
.manifestFile = pgFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
// Delta but pgFile does not match size passed, prior checksum, no compression, no pageChecksum, delta, no hasReference
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(
backupFile(repoFile, compressTypeNone, 1, true, cipherTypeAes256Cbc, STRDEF(TEST_CIPHER_PASS), fileList), 0),
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
"pg and repo file exists, pgFileMatch false, no ignoreMissing, no pageChecksum, delta, no hasReference");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.copySize, 8, "copy size set");
TEST_RESULT_UINT(result.repoSize, 32, "repo size set");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultCopy, "copy file");
TEST_RESULT_STR_Z(result.copyChecksum, "acc972a8319d4903b839c64ec217faa3e77b4fcb", "copy checksum for size passed");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum NULL");
TEST_STORAGE_GET(
storageRepo(), strZ(backupPathFile), "atestfil", .cipherType = cipherTypeAes256Cbc,
.comment = "delta, copy file (size missmatch) to encrypted repo success");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("no delta, recopy (size mismatch) file to encrypted repo");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 9,
.pgFileCopyExactSize = true,
.pgFileChecksum = STRDEF("9bc8ab2dda60ef4beed07d1e19ce0676d5edde67"),
.pgFileChecksumPage = false,
.manifestFile = pgFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(
backupFile(repoFile, compressTypeNone, 0, false, cipherTypeAes256Cbc, STRDEF(TEST_CIPHER_PASS), fileList), 0),
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
"pg and repo file exists, checksum mismatch, no ignoreMissing, no pageChecksum, no delta, no hasReference");
TEST_RESULT_UINT(result.copySize, 9, "copy size set");
TEST_RESULT_UINT(result.repoSize, 32, "repo size set");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultReCopy, "recopy file");
TEST_RESULT_STR_Z(result.copyChecksum, "9bc8ab2dda60ef4beed07d1e19ce0676d5edde67", "copy checksum");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum NULL");
TEST_STORAGE_GET(
storageRepoWrite(), strZ(backupPathFile), "atestfile", .cipherType = cipherTypeAes256Cbc,
.comment = "recopy file to encrypted repo success");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
// -------------------------------------------------------------------------------------------------------------------------
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_TITLE("no delta, recopy (checksum mismatch), file to encrypted repo");
Add local process shim. Run the local process inside a forked child process instead of exec'ing it. This allows coverage to accumulate in the local process rather than needing to test the local protocol functions directly, resulting in better end-to-end testing and less test duplication. Another advantage is that the pgbackrest binary does not need to be built for the test. The backup, restore, and verify command tests have been updated to use the new shim for coverage.
2021-05-21 12:45:00 -04:00
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
fileList = lstNewP(sizeof(BackupFile));
file = (BackupFile)
{
.pgFile = pgFile,
.pgFileIgnoreMissing = false,
.pgFileSize = 9,
.pgFileCopyExactSize = true,
.pgFileChecksum = STRDEF("1234567890123456789012345678901234567890"),
.pgFileChecksumPage = false,
.manifestFile = pgFile,
.manifestFileHasReference = false,
};
lstAdd(fileList, &file);
Add local process shim. Run the local process inside a forked child process instead of exec'ing it. This allows coverage to accumulate in the local process rather than needing to test the local protocol functions directly, resulting in better end-to-end testing and less test duplication. Another advantage is that the pgbackrest binary does not need to be built for the test. The backup, restore, and verify command tests have been updated to use the new shim for coverage.
2021-05-21 12:45:00 -04:00
TEST_ASSIGN(
result,
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
*(BackupFileResult *)lstGet(
backupFile(repoFile, compressTypeNone, 0, false, cipherTypeAes256Cbc, STRDEF(TEST_CIPHER_PASS), fileList), 0),
Add local process shim. Run the local process inside a forked child process instead of exec'ing it. This allows coverage to accumulate in the local process rather than needing to test the local protocol functions directly, resulting in better end-to-end testing and less test duplication. Another advantage is that the pgbackrest binary does not need to be built for the test. The backup, restore, and verify command tests have been updated to use the new shim for coverage.
2021-05-21 12:45:00 -04:00
"backup file");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_UINT(result.copySize, 9, "copy size set");
TEST_RESULT_UINT(result.repoSize, 32, "repo size set");
TEST_RESULT_UINT(result.backupCopyResult, backupCopyResultReCopy, "recopy file");
TEST_RESULT_STR_Z(result.copyChecksum, "9bc8ab2dda60ef4beed07d1e19ce0676d5edde67", "copy checksum for size passed");
TEST_RESULT_PTR(result.pageChecksumResult, NULL, "page checksum NULL");
TEST_STORAGE_GET(
storageRepo(), strZ(backupPathFile), "atestfile",
.cipherType = cipherTypeAes256Cbc, .comment = "recopy file to encrypted repo, success");
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
}
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// *****************************************************************************************************************************
if (testBegin("backupLabelCreate()"))
{
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
time_t timestamp = 1575401652;
String *backupLabel = backupLabelFormat(backupTypeFull, NULL, timestamp);
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("assign label when no history");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PATH_CREATE(storageRepoWrite(), STORAGE_REPO_BACKUP "/backup.history/2019");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Change meaning of TEST_RESULT_STR() macro. This macro was created before the String object existed so subsequent usage with String always included a lot of strPtr() wrapping. TEST_RESULT_STR_Z() had already been introduced but a wholesale replacement of TEST_RESULT_STR() was not done since the priority was on the C migration. Update all calls to (old) TEST_RESULT_STR() with one of the following variants: (new) TEST_RESULT_STR(), TEST_RESULT_STR_Z(), TEST_RESULT_Z(), TEST_RESULT_Z_STR().
2019-12-26 18:08:27 -07:00
TEST_RESULT_STR(backupLabelCreate(backupTypeFull, NULL, timestamp), backupLabel, "create label");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("assign label when history is older");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_EMPTY(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(),
zNewFmt(
STORAGE_REPO_BACKUP "/backup.history/2019/%s.manifest.gz",
strZ(backupLabelFormat(backupTypeFull, NULL, timestamp - 4))));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Change meaning of TEST_RESULT_STR() macro. This macro was created before the String object existed so subsequent usage with String always included a lot of strPtr() wrapping. TEST_RESULT_STR_Z() had already been introduced but a wholesale replacement of TEST_RESULT_STR() was not done since the priority was on the C migration. Update all calls to (old) TEST_RESULT_STR() with one of the following variants: (new) TEST_RESULT_STR(), TEST_RESULT_STR_Z(), TEST_RESULT_Z(), TEST_RESULT_Z_STR().
2019-12-26 18:08:27 -07:00
TEST_RESULT_STR(backupLabelCreate(backupTypeFull, NULL, timestamp), backupLabel, "create label");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("assign label when backup is older");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_EMPTY(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s", strZ(backupLabelFormat(backupTypeFull, NULL, timestamp - 2))));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Change meaning of TEST_RESULT_STR() macro. This macro was created before the String object existed so subsequent usage with String always included a lot of strPtr() wrapping. TEST_RESULT_STR_Z() had already been introduced but a wholesale replacement of TEST_RESULT_STR() was not done since the priority was on the C migration. Update all calls to (old) TEST_RESULT_STR() with one of the following variants: (new) TEST_RESULT_STR(), TEST_RESULT_STR_Z(), TEST_RESULT_Z(), TEST_RESULT_Z_STR().
2019-12-26 18:08:27 -07:00
TEST_RESULT_STR(backupLabelCreate(backupTypeFull, NULL, timestamp), backupLabel, "create label");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("advance time when backup is same");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_EMPTY(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s", strZ(backupLabelFormat(backupTypeFull, NULL, timestamp))));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_STR_Z(backupLabelCreate(backupTypeFull, NULL, timestamp), "20191203-193413F", "create label");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("error when new label is in the past even with advanced time");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_EMPTY(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s", strZ(backupLabelFormat(backupTypeFull, NULL, timestamp + 1))));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_ERROR(
Add ClockError for unexpected clock skew and timezone changes. A distinct result code should help debugging of clock skew and timezone issues.
2022-05-05 10:19:11 -04:00
backupLabelCreate(backupTypeFull, NULL, timestamp), ClockError,
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"new backup label '20191203-193413F' is not later than latest backup label '20191203-193413F'\n"
"HINT: has the timezone changed?\n"
"HINT: is there clock skew?");
Strip extensions from history manifest before showing in error message. In cases where clock skew or timezone issues are preventing backup label generation the user could see an error like this: new backup label '20220504-152308F' is not later than latest backup label '20220504-222042F_20220504-222141I.manifest.gz' This will happen if the most recent label is drawn from the history. It is cleaner (and probably less confusing) to strip off the extensions so the user sees: new backup label '20220504-152308F' is not later than latest backup label '20220504-222042F_20220504-222141I'
2022-05-05 09:20:49 -04:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("error when new label is in the past even with advanced time (from history)");
HRN_STORAGE_PUT_EMPTY(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(),
zNewFmt(
STORAGE_REPO_BACKUP "/backup.history/2019/%s.manifest.gz",
strZ(backupLabelFormat(backupTypeFull, NULL, timestamp + 3600))));
Strip extensions from history manifest before showing in error message. In cases where clock skew or timezone issues are preventing backup label generation the user could see an error like this: new backup label '20220504-152308F' is not later than latest backup label '20220504-222042F_20220504-222141I.manifest.gz' This will happen if the most recent label is drawn from the history. It is cleaner (and probably less confusing) to strip off the extensions so the user sees: new backup label '20220504-152308F' is not later than latest backup label '20220504-222042F_20220504-222141I'
2022-05-05 09:20:49 -04:00
TEST_ERROR(
Add ClockError for unexpected clock skew and timezone changes. A distinct result code should help debugging of clock skew and timezone issues.
2022-05-05 10:19:11 -04:00
backupLabelCreate(backupTypeFull, NULL, timestamp), ClockError,
Strip extensions from history manifest before showing in error message. In cases where clock skew or timezone issues are preventing backup label generation the user could see an error like this: new backup label '20220504-152308F' is not later than latest backup label '20220504-222042F_20220504-222141I.manifest.gz' This will happen if the most recent label is drawn from the history. It is cleaner (and probably less confusing) to strip off the extensions so the user sees: new backup label '20220504-152308F' is not later than latest backup label '20220504-222042F_20220504-222141I'
2022-05-05 09:20:49 -04:00
"new backup label '20191203-193413F' is not later than latest backup label '20191203-203412F'\n"
"HINT: has the timezone changed?\n"
"HINT: is there clock skew?");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// *****************************************************************************************************************************
if (testBegin("backupInit()"))
{
// Set log level to detail
harnessLogLevelSet(logLevelDetail);
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("error when backup from standby is not supported");
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptBackupStandby, true);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_ERROR(
Autogenerate test system identifiers. hrnPgControlToBuffer() and hrnPgWalToBuffer() now generate the system id based on the version of Postgres. If a value less than 100 is specified for systemId then it will be added to the default system id so there can be multiple ids for a single version of PostgreSQL. Add constants to represent version system ids in tests. These will eventually be auto-generated. This changes some checksums and we no longer have big-endian tests systems, so X those checksums out so it is obvious they are no longer valid.
2021-11-30 08:28:36 -05:00
backupInit(infoBackupNew(PG_VERSION_91, HRN_PG_SYSTEMID_91, hrnPgCatalogVersion(PG_VERSION_91), NULL)),
ConfigError, "option 'backup-standby' not valid for PostgreSQL < 9.2");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("warn and reset when backup from standby used in offline mode");
// Create pg_control
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_92);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptBackupStandby, true);
hrnCfgArgRawBool(argList, cfgOptOnline, false);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
TEST_RESULT_VOID(
Autogenerate test system identifiers. hrnPgControlToBuffer() and hrnPgWalToBuffer() now generate the system id based on the version of Postgres. If a value less than 100 is specified for systemId then it will be added to the default system id so there can be multiple ids for a single version of PostgreSQL. Add constants to represent version system ids in tests. These will eventually be auto-generated. This changes some checksums and we no longer have big-endian tests systems, so X those checksums out so it is obvious they are no longer valid.
2021-11-30 08:28:36 -05:00
backupInit(infoBackupNew(PG_VERSION_92, HRN_PG_SYSTEMID_92, hrnPgCatalogVersion(PG_VERSION_92), NULL)),
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
"backup init");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_BOOL(cfgOptionBool(cfgOptBackupStandby), false, "check backup-standby");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_LOG(
"P00 WARN: option backup-standby is enabled but backup is offline - backups will be performed from the primary");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("error when pg_control does not match stanza");
// Create pg_control
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_10);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptOnline, false);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_ERROR(
Autogenerate test system identifiers. hrnPgControlToBuffer() and hrnPgWalToBuffer() now generate the system id based on the version of Postgres. If a value less than 100 is specified for systemId then it will be added to the default system id so there can be multiple ids for a single version of PostgreSQL. Add constants to represent version system ids in tests. These will eventually be auto-generated. This changes some checksums and we no longer have big-endian tests systems, so X those checksums out so it is obvious they are no longer valid.
2021-11-30 08:28:36 -05:00
backupInit(infoBackupNew(PG_VERSION_11, HRN_PG_SYSTEMID_11, hrnPgCatalogVersion(PG_VERSION_11), NULL)),
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
BackupMismatchError,
Autogenerate test system identifiers. hrnPgControlToBuffer() and hrnPgWalToBuffer() now generate the system id based on the version of Postgres. If a value less than 100 is specified for systemId then it will be added to the default system id so there can be multiple ids for a single version of PostgreSQL. Add constants to represent version system ids in tests. These will eventually be auto-generated. This changes some checksums and we no longer have big-endian tests systems, so X those checksums out so it is obvious they are no longer valid.
2021-11-30 08:28:36 -05:00
"PostgreSQL version 10, system-id " HRN_PG_SYSTEMID_10_Z " do not match stanza version 11, system-id"
" " HRN_PG_SYSTEMID_11_Z "\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"HINT: is this the correct stanza?");
TEST_ERROR(
Autogenerate test system identifiers. hrnPgControlToBuffer() and hrnPgWalToBuffer() now generate the system id based on the version of Postgres. If a value less than 100 is specified for systemId then it will be added to the default system id so there can be multiple ids for a single version of PostgreSQL. Add constants to represent version system ids in tests. These will eventually be auto-generated. This changes some checksums and we no longer have big-endian tests systems, so X those checksums out so it is obvious they are no longer valid.
2021-11-30 08:28:36 -05:00
backupInit(infoBackupNew(PG_VERSION_10, HRN_PG_SYSTEMID_11, hrnPgCatalogVersion(PG_VERSION_10), NULL)),
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
BackupMismatchError,
Autogenerate test system identifiers. hrnPgControlToBuffer() and hrnPgWalToBuffer() now generate the system id based on the version of Postgres. If a value less than 100 is specified for systemId then it will be added to the default system id so there can be multiple ids for a single version of PostgreSQL. Add constants to represent version system ids in tests. These will eventually be auto-generated. This changes some checksums and we no longer have big-endian tests systems, so X those checksums out so it is obvious they are no longer valid.
2021-11-30 08:28:36 -05:00
"PostgreSQL version 10, system-id " HRN_PG_SYSTEMID_10_Z " do not match stanza version 10, system-id"
" " HRN_PG_SYSTEMID_11_Z "\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"HINT: is this the correct stanza?");
// -------------------------------------------------------------------------------------------------------------------------
Don't warn when stop-auto is enabled on PostgreSQL >= 9.6. PostgreSQL >= 9.6 uses non-exclusive backup which has implicit stop-auto since the backup will stop when the connection is terminated. The warning was made more verbose in 1f2ce45e but this now seems like a bad idea since there are likely users with mixed version environments where stop-auto is enabled globally. There's no reason to fill their logs with warnings over a harmless option. If anything we should warn when stop-auto is explicitly set to false but this doesn't seem very important either. Revert to the prior behavior, which is to warn and reset when stop-auto is enabled on PostgreSQL < 9.3.
2019-12-14 09:53:50 -05:00
TEST_TITLE("reset stop-auto when PostgreSQL < 9.3");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Create pg_control
Remove support for PostgreSQL 8.3/8.4. There is no evidence that users need 8.3/8.4 anymore but it does cost us in terms of development and testing, especially now that we have a number of new backup/restore features planned. It seems to make sense to remove this support now. If there are users who need to use/migrate from these versions they can use an older version of pgBackRest.
2022-01-06 15:34:04 -05:00
HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_90);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptOnline, false);
hrnCfgArgRawBool(argList, cfgOptStopAuto, true);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
TEST_RESULT_VOID(
Remove support for PostgreSQL 8.3/8.4. There is no evidence that users need 8.3/8.4 anymore but it does cost us in terms of development and testing, especially now that we have a number of new backup/restore features planned. It seems to make sense to remove this support now. If there are users who need to use/migrate from these versions they can use an older version of pgBackRest.
2022-01-06 15:34:04 -05:00
backupInit(infoBackupNew(PG_VERSION_90, HRN_PG_SYSTEMID_90, hrnPgCatalogVersion(PG_VERSION_90), NULL)),
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
"backup init");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_BOOL(cfgOptionBool(cfgOptStopAuto), false, "check stop-auto");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Don't warn when stop-auto is enabled on PostgreSQL >= 9.6. PostgreSQL >= 9.6 uses non-exclusive backup which has implicit stop-auto since the backup will stop when the connection is terminated. The warning was made more verbose in 1f2ce45e but this now seems like a bad idea since there are likely users with mixed version environments where stop-auto is enabled globally. There's no reason to fill their logs with warnings over a harmless option. If anything we should warn when stop-auto is explicitly set to false but this doesn't seem very important either. Revert to the prior behavior, which is to warn and reset when stop-auto is enabled on PostgreSQL < 9.3.
2019-12-14 09:53:50 -05:00
TEST_RESULT_LOG("P00 WARN: stop-auto option is only available in PostgreSQL >= 9.3");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("reset checksum-page when the cluster does not have checksums enabled");
// Create pg_control
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_93);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Create stanza
argList = strLstNew();
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
hrnCfgArgRawBool(argList, cfgOptOnline, false);
HRN_CFG_LOAD(cfgCmdStanzaCreate, argList);
cmdStanzaCreate();
TEST_RESULT_LOG("P00 INFO: stanza-create for stanza 'test1' on repo1");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptChecksumPage, true);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
harnessPqScriptSet((HarnessPq [])
{
// Connect to primary
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRNPQ_MACRO_OPEN_GE_92(1, "dbname='postgres' port=5432", PG_VERSION_93, TEST_PATH "/pg1", false, NULL, NULL),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
HRNPQ_MACRO_DONE()
});
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
TEST_RESULT_VOID(
Autogenerate test system identifiers. hrnPgControlToBuffer() and hrnPgWalToBuffer() now generate the system id based on the version of Postgres. If a value less than 100 is specified for systemId then it will be added to the default system id so there can be multiple ids for a single version of PostgreSQL. Add constants to represent version system ids in tests. These will eventually be auto-generated. This changes some checksums and we no longer have big-endian tests systems, so X those checksums out so it is obvious they are no longer valid.
2021-11-30 08:28:36 -05:00
dbFree(
backupInit(infoBackupNew(PG_VERSION_93, HRN_PG_SYSTEMID_93, hrnPgCatalogVersion(PG_VERSION_93), NULL))->dbPrimary),
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
"backup init");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_BOOL(cfgOptionBool(cfgOptChecksumPage), false, "check checksum-page");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_LOG(
"P00 WARN: checksum-page option set to true but checksums are not enabled on the cluster, resetting to false");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("ok if cluster checksums are enabled and checksum-page is any value");
// Create pg_control with page checksums
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_93, .pageChecksum = true);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptChecksumPage, false);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
harnessPqScriptSet((HarnessPq [])
{
// Connect to primary
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRNPQ_MACRO_OPEN_GE_92(1, "dbname='postgres' port=5432", PG_VERSION_93, TEST_PATH "/pg1", false, NULL, NULL),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
HRNPQ_MACRO_DONE()
});
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
TEST_RESULT_VOID(
Autogenerate test system identifiers. hrnPgControlToBuffer() and hrnPgWalToBuffer() now generate the system id based on the version of Postgres. If a value less than 100 is specified for systemId then it will be added to the default system id so there can be multiple ids for a single version of PostgreSQL. Add constants to represent version system ids in tests. These will eventually be auto-generated. This changes some checksums and we no longer have big-endian tests systems, so X those checksums out so it is obvious they are no longer valid.
2021-11-30 08:28:36 -05:00
dbFree(
backupInit(infoBackupNew(PG_VERSION_93, HRN_PG_SYSTEMID_93, hrnPgCatalogVersion(PG_VERSION_93), NULL))->dbPrimary),
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
"backup init");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_BOOL(cfgOptionBool(cfgOptChecksumPage), false, "check checksum-page");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Create pg_control without page checksums
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_93);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
harnessPqScriptSet((HarnessPq [])
{
// Connect to primary
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRNPQ_MACRO_OPEN_GE_92(1, "dbname='postgres' port=5432", PG_VERSION_93, TEST_PATH "/pg1", false, NULL, NULL),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
HRNPQ_MACRO_DONE()
});
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
TEST_RESULT_VOID(
Autogenerate test system identifiers. hrnPgControlToBuffer() and hrnPgWalToBuffer() now generate the system id based on the version of Postgres. If a value less than 100 is specified for systemId then it will be added to the default system id so there can be multiple ids for a single version of PostgreSQL. Add constants to represent version system ids in tests. These will eventually be auto-generated. This changes some checksums and we no longer have big-endian tests systems, so X those checksums out so it is obvious they are no longer valid.
2021-11-30 08:28:36 -05:00
dbFree(
backupInit(infoBackupNew(PG_VERSION_93, HRN_PG_SYSTEMID_93, hrnPgCatalogVersion(PG_VERSION_93), NULL))->dbPrimary),
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
"backup init");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_BOOL(cfgOptionBool(cfgOptChecksumPage), false, "check checksum-page");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// *****************************************************************************************************************************
if (testBegin("backupTime()"))
{
// -------------------------------------------------------------------------------------------------------------------------
Add retries to PostgreSQL sleep when starting a backup. Inaccuracies in sleep time or clock skew might make a single sleep insufficient to reach the next second. Add a few retries to make the process more reliable but still avoid an infinite loop if something is seriously wrong.
2020-12-02 22:41:14 -05:00
TEST_TITLE("sleep retries and stall error");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Create pg_control
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_93);
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Create stanza
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
hrnCfgArgRawBool(argList, cfgOptOnline, false);
HRN_CFG_LOAD(cfgCmdStanzaCreate, argList);
cmdStanzaCreate();
TEST_RESULT_LOG("P00 INFO: stanza-create for stanza 'test1' on repo1");
argList = strLstNew();
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
harnessPqScriptSet((HarnessPq [])
{
// Connect to primary
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRNPQ_MACRO_OPEN_GE_92(1, "dbname='postgres' port=5432", PG_VERSION_93, TEST_PATH "/pg1", false, NULL, NULL),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Add retries to PostgreSQL sleep when starting a backup. Inaccuracies in sleep time or clock skew might make a single sleep insufficient to reach the next second. Add a few retries to make the process more reliable but still avoid an infinite loop if something is seriously wrong.
2020-12-02 22:41:14 -05:00
// Advance the time slowly to force retries
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
HRNPQ_MACRO_TIME_QUERY(1, 1575392588998),
HRNPQ_MACRO_TIME_QUERY(1, 1575392588999),
Add retries to PostgreSQL sleep when starting a backup. Inaccuracies in sleep time or clock skew might make a single sleep insufficient to reach the next second. Add a few retries to make the process more reliable but still avoid an infinite loop if something is seriously wrong.
2020-12-02 22:41:14 -05:00
HRNPQ_MACRO_TIME_QUERY(1, 1575392589001),
// Stall time to force an error
HRNPQ_MACRO_TIME_QUERY(1, 1575392589998),
HRNPQ_MACRO_TIME_QUERY(1, 1575392589997),
HRNPQ_MACRO_TIME_QUERY(1, 1575392589998),
HRNPQ_MACRO_TIME_QUERY(1, 1575392589999),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
HRNPQ_MACRO_DONE()
});
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
BackupData *backupData = backupInit(
Autogenerate test system identifiers. hrnPgControlToBuffer() and hrnPgWalToBuffer() now generate the system id based on the version of Postgres. If a value less than 100 is specified for systemId then it will be added to the default system id so there can be multiple ids for a single version of PostgreSQL. Add constants to represent version system ids in tests. These will eventually be auto-generated. This changes some checksums and we no longer have big-endian tests systems, so X those checksums out so it is obvious they are no longer valid.
2021-11-30 08:28:36 -05:00
infoBackupNew(PG_VERSION_93, HRN_PG_SYSTEMID_93, hrnPgCatalogVersion(PG_VERSION_93), NULL));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Add retries to PostgreSQL sleep when starting a backup. Inaccuracies in sleep time or clock skew might make a single sleep insufficient to reach the next second. Add a few retries to make the process more reliable but still avoid an infinite loop if something is seriously wrong.
2020-12-02 22:41:14 -05:00
TEST_RESULT_INT(backupTime(backupData, true), 1575392588, "multiple tries for sleep");
TEST_ERROR(backupTime(backupData, true), KernelError, "PostgreSQL clock has not advanced to the next second after 3 tries");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
dbFree(backupData->dbPrimary);
}
// *****************************************************************************************************************************
if (testBegin("backupResumeFind()"))
{
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptPgPath, "/pg");
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Convert BackupType enum to StringId. Allows removal of backupType()/backupTypeStr() and improves debug logging of the enum. Move BackupType enum and string constants to info/infoBackup.h so they are available to more modules. Also convert InfoBackup to use BackupType instead of a String.
2021-05-03 12:15:39 -04:00
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeFull);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptCompress, false);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
Fix resume after partial delete of backup by prior resume. If files other than backup.manifest.copy were left in a backup path by a prior resume then the next resume would skip the backup rather than removing it. Since the backup path still existed, it would be found during backup label generation and cause an error if it appeared to be later than the new backup label. This occurred if the skipped backup was full. The error was only likely on object stores such as S3 because of the order of file deletion. Posix file systems delete from the bottom up because directories containing files cannot be deleted. Object stores do not have directories so files are deleted in whatever order they are provided by the list command. However, the issue can be reproduced on a Posix file system by manually deleting backup.manifest.copy from a resumable backup path. Fix the issue by removing the resumable backup if it has no manifest files. Also add a new warning message for this condition. Note that this issue could be resolved by running expire or a new full backup.
2021-01-12 12:38:32 -05:00
TEST_TITLE("cannot resume when manifest and copy are missing");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PATH_CREATE(storageRepoWrite(), STORAGE_REPO_BACKUP "/20191003-105320F");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_PTR(backupResumeFind((Manifest *)1, NULL), NULL, "find resumable backup");
Fix resume after partial delete of backup by prior resume. If files other than backup.manifest.copy were left in a backup path by a prior resume then the next resume would skip the backup rather than removing it. Since the backup path still existed, it would be found during backup label generation and cause an error if it appeared to be later than the new backup label. This occurred if the skipped backup was full. The error was only likely on object stores such as S3 because of the order of file deletion. Posix file systems delete from the bottom up because directories containing files cannot be deleted. Object stores do not have directories so files are deleted in whatever order they are provided by the list command. However, the issue can be reproduced on a Posix file system by manually deleting backup.manifest.copy from a resumable backup path. Fix the issue by removing the resumable backup if it has no manifest files. Also add a new warning message for this condition. Note that this issue could be resolved by running expire or a new full backup.
2021-01-12 12:38:32 -05:00
TEST_RESULT_LOG(
"P00 WARN: backup '20191003-105320F' cannot be resumed: partially deleted by prior resume or invalid");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("cannot resume when resume is disabled");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PATH_CREATE(storageRepoWrite(), STORAGE_REPO_BACKUP "/20191003-105320F");
Fix resume after partial delete of backup by prior resume. If files other than backup.manifest.copy were left in a backup path by a prior resume then the next resume would skip the backup rather than removing it. Since the backup path still existed, it would be found during backup label generation and cause an error if it appeared to be later than the new backup label. This occurred if the skipped backup was full. The error was only likely on object stores such as S3 because of the order of file deletion. Posix file systems delete from the bottom up because directories containing files cannot be deleted. Object stores do not have directories so files are deleted in whatever order they are provided by the list command. However, the issue can be reproduced on a Posix file system by manually deleting backup.manifest.copy from a resumable backup path. Fix the issue by removing the resumable backup if it has no manifest files. Also add a new warning message for this condition. Note that this issue could be resolved by running expire or a new full backup.
2021-01-12 12:38:32 -05:00
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
cfgOptionSet(cfgOptResume, cfgSourceParam, BOOL_FALSE_VAR);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_EMPTY(storageRepoWrite(), STORAGE_REPO_BACKUP "/20191003-105320F/" BACKUP_MANIFEST_FILE INFO_COPY_EXT);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_PTR(backupResumeFind((Manifest *)1, NULL), NULL, "find resumable backup");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
TEST_RESULT_LOG("P00 INFO: backup '20191003-105320F' cannot be resumed: resume is disabled");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_STORAGE_LIST_EMPTY(storageRepo(), STORAGE_REPO_BACKUP, .comment = "check backup path removed");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
cfgOptionSet(cfgOptResume, cfgSourceParam, BOOL_TRUE_VAR);
Add coverage for empty CATCH() blocks. Currently empty CATCH() blocks are always marked as covered because of the loop structure of error handling. A prototype implementation of error handling without looping has shown that these CATCH() blocks are not covered without new tests. Whether or not that prototype gets committed it is worth adding the tests.
2021-10-26 13:53:44 -04:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("cannot resume when error on manifest load");
Manifest *manifest = NULL;
Improve memory usage of mem contexts. Each mem context can track child contexts, allocations, and a callback. Before this change memory was allocated for tracking all three even if they were not used for a particular context. This made mem contexts unsuitable for String and Variant objects since they are plentiful and need to be as small as possible. This change allows mem contexts to be configured to track any combination of child contexts, allocations, and a callback. In addition, the mem context can be configured to track a single child context and/or allocation, which saves memory and is a common use case. Another benefit is that Variants can own objects (e.g. KeyValue) that they encapsulate. All of this makes memory accounting simpler because mem contexts have names while allocations do not. No more memory is used than before since Variants and Strings still had to store the memory context they were originally allocated in so they could be easily freed. Update the String and Variant objects to use this new functionality. The custom strFree() and varFree() functions are no longer required and can now be a wrapper around objFree(). Lastly, this will allow strMove() and varMove() to be implemented and used in cases where strDup() and varDup() are being used to move a String or Variant to a new context. Since this will be a bit noisy it is saved for a future commit.
2022-05-18 10:52:01 -04:00
OBJ_NEW_BEGIN(Manifest, .childQty = MEM_CONTEXT_QTY_MAX)
Add coverage for empty CATCH() blocks. Currently empty CATCH() blocks are always marked as covered because of the loop structure of error handling. A prototype implementation of error handling without looping has shown that these CATCH() blocks are not covered without new tests. Whether or not that prototype gets committed it is worth adding the tests.
2021-10-26 13:53:44 -04:00
{
manifest = manifestNewInternal();
manifest->pub.data.backupType = backupTypeFull;
manifest->pub.data.backrestVersion = STRDEF("BOGUS");
}
OBJ_NEW_END();
HRN_STORAGE_PUT_Z(storageRepoWrite(), STORAGE_REPO_BACKUP "/20191003-105320F/" BACKUP_MANIFEST_FILE INFO_COPY_EXT, "X");
TEST_RESULT_PTR(backupResumeFind(manifest, NULL), NULL, "find resumable backup");
TEST_RESULT_LOG(
"P00 WARN: backup '20191003-105320F' cannot be resumed: unable to read"
" <REPO:BACKUP>/20191003-105320F/backup.manifest.copy");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("cannot resume when pgBackRest version has changed");
Allow additional memory to be allocated with a mem context. The primary benefit is that objects can allocate memory for their struct with the context, which saves an additional allocation and makes it easier to read context/allocation dumps. Also, the memory context does not need to be stored with the object since it can be determined using the object pointer. Object pointers cannot be moved, so this means whatever additional memory is allocated cannot be resized. That makes the additional memory ideal for object structs, but not so much for allocating a list that might change size. Mem contexts can no longer be reused since they will probably be the wrong size so their memory is freed on memContextFree(). This still means fewer allocations and frees overall. Interfaces still need to be freed by mem context so the old objMove() and objFree() have been preserved as objMoveContext() and objFreeContext(). This will be addressed in a future commit.
2021-09-01 11:10:35 -04:00
Manifest *manifestResume = NULL;
Improve memory usage of mem contexts. Each mem context can track child contexts, allocations, and a callback. Before this change memory was allocated for tracking all three even if they were not used for a particular context. This made mem contexts unsuitable for String and Variant objects since they are plentiful and need to be as small as possible. This change allows mem contexts to be configured to track any combination of child contexts, allocations, and a callback. In addition, the mem context can be configured to track a single child context and/or allocation, which saves memory and is a common use case. Another benefit is that Variants can own objects (e.g. KeyValue) that they encapsulate. All of this makes memory accounting simpler because mem contexts have names while allocations do not. No more memory is used than before since Variants and Strings still had to store the memory context they were originally allocated in so they could be easily freed. Update the String and Variant objects to use this new functionality. The custom strFree() and varFree() functions are no longer required and can now be a wrapper around objFree(). Lastly, this will allow strMove() and varMove() to be implemented and used in cases where strDup() and varDup() are being used to move a String or Variant to a new context. Since this will be a bit noisy it is saved for a future commit.
2022-05-18 10:52:01 -04:00
OBJ_NEW_BEGIN(Manifest, .childQty = MEM_CONTEXT_QTY_MAX)
Allow additional memory to be allocated with a mem context. The primary benefit is that objects can allocate memory for their struct with the context, which saves an additional allocation and makes it easier to read context/allocation dumps. Also, the memory context does not need to be stored with the object since it can be determined using the object pointer. Object pointers cannot be moved, so this means whatever additional memory is allocated cannot be resized. That makes the additional memory ideal for object structs, but not so much for allocating a list that might change size. Mem contexts can no longer be reused since they will probably be the wrong size so their memory is freed on memContextFree(). This still means fewer allocations and frees overall. Interfaces still need to be freed by mem context so the old objMove() and objFree() have been preserved as objMoveContext() and objFreeContext(). This will be addressed in a future commit.
2021-09-01 11:10:35 -04:00
{
manifestResume = manifestNewInternal();
manifestResume->pub.info = infoNew(NULL);
manifestResume->pub.data.backupType = backupTypeFull;
Use strNewZ() in cases where STRDEF() assignment goes out of scope. If a variable assigned with STRDEF() is referenced out of scope of the STRDEF() assignment then the value is undefined. Luckily most of the instances are in tests but there is one in the core code. It is not clear if this is a live bug or not but it certainly needs to be fixed. Found with -fsanitize=address.
2022-03-24 12:26:09 -06:00
manifestResume->pub.data.backupLabel = strNewZ("20191003-105320F");
Allow additional memory to be allocated with a mem context. The primary benefit is that objects can allocate memory for their struct with the context, which saves an additional allocation and makes it easier to read context/allocation dumps. Also, the memory context does not need to be stored with the object since it can be determined using the object pointer. Object pointers cannot be moved, so this means whatever additional memory is allocated cannot be resized. That makes the additional memory ideal for object structs, but not so much for allocating a list that might change size. Mem contexts can no longer be reused since they will probably be the wrong size so their memory is freed on memContextFree(). This still means fewer allocations and frees overall. Interfaces still need to be freed by mem context so the old objMove() and objFree() have been preserved as objMoveContext() and objFreeContext(). This will be addressed in a future commit.
2021-09-01 11:10:35 -04:00
manifestResume->pub.data.pgVersion = PG_VERSION_12;
}
OBJ_NEW_END();
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
manifestTargetAdd(manifestResume, &(ManifestTarget){.name = MANIFEST_TARGET_PGDATA_STR, .path = STRDEF("/pg")});
manifestPathAdd(manifestResume, &(ManifestPath){.name = MANIFEST_TARGET_PGDATA_STR});
Pass file by reference in manifestFileAdd(). Coverity complained that this pass by value was inefficient: CID 376402: Performance inefficiencies (PASS_BY_VALUE) Passing parameter file of type "ManifestFile" (size 136 bytes) by value. This was completely intentional since it gives us a copy of the struct that we can change without bothering the caller. However, updating fields is fine and may benefit the caller at some future data, and in any case does no harm now. And as usual it is easier not to fight with Coverity.
2022-02-20 16:45:07 -06:00
manifestFileAdd(manifestResume, &(ManifestFile){.name = STRDEF("pg_data/" PG_FILE_PGVERSION)});
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
manifestSave(
manifestResume,
storageWriteIo(
storageNewWriteP(
storageRepoWrite(), STRDEF(STORAGE_REPO_BACKUP "/20191003-105320F/" BACKUP_MANIFEST_FILE INFO_COPY_EXT))));
TEST_RESULT_PTR(backupResumeFind(manifest, NULL), NULL, "find resumable backup");
TEST_RESULT_LOG(
"P00 WARN: backup '20191003-105320F' cannot be resumed:"
" new pgBackRest version 'BOGUS' does not match resumable pgBackRest version '" PROJECT_VERSION "'");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_STORAGE_LIST_EMPTY(storageRepo(), STORAGE_REPO_BACKUP, .comment = "check backup path removed");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Refactor info modules with inline getters/setters. Extend the pattern introduced in 79a2d02c to the info modules.
2021-04-09 13:48:40 -04:00
manifest->pub.data.backrestVersion = STRDEF(PROJECT_VERSION);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("cannot resume when backup labels do not match (resumable is null)");
Refactor info modules with inline getters/setters. Extend the pattern introduced in 79a2d02c to the info modules.
2021-04-09 13:48:40 -04:00
manifest->pub.data.backupType = backupTypeFull;
manifest->pub.data.backupLabelPrior = STRDEF("20191003-105320F");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
manifestSave(
manifestResume,
storageWriteIo(
storageNewWriteP(
storageRepoWrite(), STRDEF(STORAGE_REPO_BACKUP "/20191003-105320F/" BACKUP_MANIFEST_FILE INFO_COPY_EXT))));
TEST_RESULT_PTR(backupResumeFind(manifest, NULL), NULL, "find resumable backup");
TEST_RESULT_LOG(
"P00 WARN: backup '20191003-105320F' cannot be resumed:"
" new prior backup label '<undef>' does not match resumable prior backup label '20191003-105320F'");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_STORAGE_LIST_EMPTY(storageRepo(), STORAGE_REPO_BACKUP, .comment = "check backup path removed");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Refactor info modules with inline getters/setters. Extend the pattern introduced in 79a2d02c to the info modules.
2021-04-09 13:48:40 -04:00
manifest->pub.data.backupLabelPrior = NULL;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("cannot resume when backup labels do not match (new is null)");
Refactor info modules with inline getters/setters. Extend the pattern introduced in 79a2d02c to the info modules.
2021-04-09 13:48:40 -04:00
manifest->pub.data.backupType = backupTypeFull;
manifestResume->pub.data.backupLabelPrior = STRDEF("20191003-105320F");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
manifestSave(
manifestResume,
storageWriteIo(
storageNewWriteP(
storageRepoWrite(), STRDEF(STORAGE_REPO_BACKUP "/20191003-105320F/" BACKUP_MANIFEST_FILE INFO_COPY_EXT))));
TEST_RESULT_PTR(backupResumeFind(manifest, NULL), NULL, "find resumable backup");
TEST_RESULT_LOG(
"P00 WARN: backup '20191003-105320F' cannot be resumed:"
" new prior backup label '20191003-105320F' does not match resumable prior backup label '<undef>'");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_STORAGE_LIST_EMPTY(storageRepo(), STORAGE_REPO_BACKUP, .comment = "check backup path removed");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Refactor info modules with inline getters/setters. Extend the pattern introduced in 79a2d02c to the info modules.
2021-04-09 13:48:40 -04:00
manifestResume->pub.data.backupLabelPrior = NULL;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("cannot resume when compression does not match");
Refactor info modules with inline getters/setters. Extend the pattern introduced in 79a2d02c to the info modules.
2021-04-09 13:48:40 -04:00
manifestResume->pub.data.backupOptionCompressType = compressTypeGz;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
manifestSave(
manifestResume,
storageWriteIo(
storageNewWriteP(
storageRepoWrite(), STRDEF(STORAGE_REPO_BACKUP "/20191003-105320F/" BACKUP_MANIFEST_FILE INFO_COPY_EXT))));
TEST_RESULT_PTR(backupResumeFind(manifest, NULL), NULL, "find resumable backup");
TEST_RESULT_LOG(
"P00 WARN: backup '20191003-105320F' cannot be resumed:"
Add infrastructure for multiple compression type support. Add compress-type option and deprecate compress option. Since the compress option is boolean it won't work with multiple compression types. Add logic to cfgLoadUpdateOption() to update compress-type if it is not set directly. The compress option should no longer be referenced outside the cfgLoadUpdateOption() function. Add common/compress/helper module to contain interface functions that work with multiple compression types. Code outside this module should no longer call specific compression drivers, though it may be OK to reference a specific compression type using the new interface (e.g., saving backup history files in gz format). Unit tests only test compression using the gz format because other formats may not be available in all builds. It is the job of integration tests to exercise all compression types. Additional compression types will be added in future commits.
2020-03-06 14:41:03 -05:00
" new compression 'none' does not match resumable compression 'gz'");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_STORAGE_LIST_EMPTY(storageRepo(), STORAGE_REPO_BACKUP, .comment = "check backup path removed");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Refactor info modules with inline getters/setters. Extend the pattern introduced in 79a2d02c to the info modules.
2021-04-09 13:48:40 -04:00
manifestResume->pub.data.backupOptionCompressType = compressTypeNone;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// *****************************************************************************************************************************
if (testBegin("backupJobResult()"))
{
// Set log level to detail
harnessLogLevelSet(logLevelDetail);
// -------------------------------------------------------------------------------------------------------------------------
Fix missing files corrupting the manifest. If a file was removed by PostgreSQL during the backup (or was missing from the standby) then the next file might not be copied and updated in the manifest. If this happened then the backup would error when restored. The issue was that removing files from the manifest invalidated the pointers stored in the processing queues. When a file was removed, all the pointers shifted to the next file in the list, causing a file to be unprocessed. Since the unprocessed file was still in the manifest it would be saved with no checksum, causing a failure on restore. When process-max was > 1 then the bug would often not express since the file had already been pulled from the queue and updates to the manifest are done by name rather than by pointer.
2020-01-26 13:19:13 -07:00
TEST_TITLE("report job error");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Simplify strIdFrom*() functions. The strIdFrom*() forced the caller to pick an encoding, which led to a number of TRY...CATCH blocks in the code. In practice the caller does not care which encoding is used as long as the string is valid for some encoding. Update the strIdFrom*() function to try all possible encodings and only throw an error when the string is not valid for any of them.
2021-11-01 10:08:56 -04:00
ProtocolParallelJob *job = protocolParallelJobNew(VARSTRDEF("key"), protocolCommandNew(strIdFromZ("x")));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
protocolParallelJobErrorSet(job, errorTypeCode(&AssertError), STRDEF("error message"));
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
unsigned int currentPercentComplete = 0;
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
TEST_ERROR(
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
backupJobResult((Manifest *)1, NULL, storageTest, strLstNew(), job, false, 0, NULL, &currentPercentComplete),
AssertError, "error message");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
Fix missing files corrupting the manifest. If a file was removed by PostgreSQL during the backup (or was missing from the standby) then the next file might not be copied and updated in the manifest. If this happened then the backup would error when restored. The issue was that removing files from the manifest invalidated the pointers stored in the processing queues. When a file was removed, all the pointers shifted to the next file in the list, causing a file to be unprocessed. Since the unprocessed file was still in the manifest it would be saved with no checksum, causing a failure on restore. When process-max was > 1 then the bug would often not express since the file had already been pulled from the queue and updates to the manifest are done by name rather than by pointer.
2020-01-26 13:19:13 -07:00
TEST_TITLE("report host/100% progress on noop result");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Create job that skips file
Simplify strIdFrom*() functions. The strIdFrom*() forced the caller to pick an encoding, which led to a number of TRY...CATCH blocks in the code. In practice the caller does not care which encoding is used as long as the string is valid for some encoding. Update the strIdFrom*() function to try all possible encodings and only throw an error when the string is not valid for any of them.
2021-11-01 10:08:56 -04:00
job = protocolParallelJobNew(VARSTRDEF("pg_data/test"), protocolCommandNew(strIdFromZ("x")));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Binary protocol. Switch from JSON-based to binary protocol for communicating with local and remote process. The pack type is used to implement the binary protocol. There are a number advantages: * The pack type is more compact than JSON and are more efficient to render/parse. * Packs are more strictly typed than JSON. * Each protocol message is written entirely within ProtocolServer/ProtocolClient so is less likely to get interrupted by an error and leave the protocol in a bad state. * There is no limit on message size. Previously this was limited by buffer size without a custom implementation, as was done for read/writing files. Some cruft from the Perl days was removed, specifically allowing NULL messages and stack traces. This is no longer possible in C. There is room for improvement here, in particular locking down the allowed sequence of protocol messages and building a state machine to enforce it. This will be useful for resetting the protocol when it gets in a bad state.
2021-06-24 13:31:16 -04:00
PackWrite *const resultPack = protocolPackNew();
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
pckWriteStrP(resultPack, STRDEF("pg_data/test"));
Binary protocol. Switch from JSON-based to binary protocol for communicating with local and remote process. The pack type is used to implement the binary protocol. There are a number advantages: * The pack type is more compact than JSON and are more efficient to render/parse. * Packs are more strictly typed than JSON. * Each protocol message is written entirely within ProtocolServer/ProtocolClient so is less likely to get interrupted by an error and leave the protocol in a bad state. * There is no limit on message size. Previously this was limited by buffer size without a custom implementation, as was done for read/writing files. Some cruft from the Perl days was removed, specifically allowing NULL messages and stack traces. This is no longer possible in C. There is room for improvement here, in particular locking down the allowed sequence of protocol messages and building a state machine to enforce it. This will be useful for resetting the protocol when it gets in a bad state.
2021-06-24 13:31:16 -04:00
pckWriteU32P(resultPack, backupCopyResultNoOp);
pckWriteU64P(resultPack, 0);
pckWriteU64P(resultPack, 0);
pckWriteStrP(resultPack, NULL);
pckWriteStrP(resultPack, NULL);
pckWriteEndP(resultPack);
Add Pack pseudo-type. Rather than working directly with Buffer types, define a new Pack pseudo-type that represents a Buffer containing a pack. This makes it clearer that a pack is being stored and allows stronger typing.
2021-09-23 08:31:32 -04:00
protocolParallelJobResultSet(job, pckReadNew(pckWriteResult(resultPack)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Create manifest with file
Allow additional memory to be allocated with a mem context. The primary benefit is that objects can allocate memory for their struct with the context, which saves an additional allocation and makes it easier to read context/allocation dumps. Also, the memory context does not need to be stored with the object since it can be determined using the object pointer. Object pointers cannot be moved, so this means whatever additional memory is allocated cannot be resized. That makes the additional memory ideal for object structs, but not so much for allocating a list that might change size. Mem contexts can no longer be reused since they will probably be the wrong size so their memory is freed on memContextFree(). This still means fewer allocations and frees overall. Interfaces still need to be freed by mem context so the old objMove() and objFree() have been preserved as objMoveContext() and objFreeContext(). This will be addressed in a future commit.
2021-09-01 11:10:35 -04:00
Manifest *manifest = NULL;
Improve memory usage of mem contexts. Each mem context can track child contexts, allocations, and a callback. Before this change memory was allocated for tracking all three even if they were not used for a particular context. This made mem contexts unsuitable for String and Variant objects since they are plentiful and need to be as small as possible. This change allows mem contexts to be configured to track any combination of child contexts, allocations, and a callback. In addition, the mem context can be configured to track a single child context and/or allocation, which saves memory and is a common use case. Another benefit is that Variants can own objects (e.g. KeyValue) that they encapsulate. All of this makes memory accounting simpler because mem contexts have names while allocations do not. No more memory is used than before since Variants and Strings still had to store the memory context they were originally allocated in so they could be easily freed. Update the String and Variant objects to use this new functionality. The custom strFree() and varFree() functions are no longer required and can now be a wrapper around objFree(). Lastly, this will allow strMove() and varMove() to be implemented and used in cases where strDup() and varDup() are being used to move a String or Variant to a new context. Since this will be a bit noisy it is saved for a future commit.
2022-05-18 10:52:01 -04:00
OBJ_NEW_BEGIN(Manifest, .childQty = MEM_CONTEXT_QTY_MAX)
Allow additional memory to be allocated with a mem context. The primary benefit is that objects can allocate memory for their struct with the context, which saves an additional allocation and makes it easier to read context/allocation dumps. Also, the memory context does not need to be stored with the object since it can be determined using the object pointer. Object pointers cannot be moved, so this means whatever additional memory is allocated cannot be resized. That makes the additional memory ideal for object structs, but not so much for allocating a list that might change size. Mem contexts can no longer be reused since they will probably be the wrong size so their memory is freed on memContextFree(). This still means fewer allocations and frees overall. Interfaces still need to be freed by mem context so the old objMove() and objFree() have been preserved as objMoveContext() and objFreeContext(). This will be addressed in a future commit.
2021-09-01 11:10:35 -04:00
{
manifest = manifestNewInternal();
Pass file by reference in manifestFileAdd(). Coverity complained that this pass by value was inefficient: CID 376402: Performance inefficiencies (PASS_BY_VALUE) Passing parameter file of type "ManifestFile" (size 136 bytes) by value. This was completely intentional since it gives us a copy of the struct that we can change without bothering the caller. However, updating fields is fine and may benefit the caller at some future data, and in any case does no harm now. And as usual it is easier not to fight with Coverity.
2022-02-20 16:45:07 -06:00
manifestFileAdd(manifest, &(ManifestFile){.name = STRDEF("pg_data/test")});
Allow additional memory to be allocated with a mem context. The primary benefit is that objects can allocate memory for their struct with the context, which saves an additional allocation and makes it easier to read context/allocation dumps. Also, the memory context does not need to be stored with the object since it can be determined using the object pointer. Object pointers cannot be moved, so this means whatever additional memory is allocated cannot be resized. That makes the additional memory ideal for object structs, but not so much for allocating a list that might change size. Mem contexts can no longer be reused since they will probably be the wrong size so their memory is freed on memContextFree(). This still means fewer allocations and frees overall. Interfaces still need to be freed by mem context so the old objMove() and objFree() have been preserved as objMoveContext() and objFreeContext(). This will be addressed in a future commit.
2021-09-01 11:10:35 -04:00
}
OBJ_NEW_END();
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Match backup log size with size reported by info command. Properly log the size of files copied during the backup, matching the backup size returned from the info command. In the reference issue, the incremental backup after switchover logs the size of all files evaluated rather than only the size of the files copied in the backup.
2021-11-09 13:24:56 -05:00
uint64_t sizeProgress = 0;
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
currentPercentComplete = 4567;
Match backup log size with size reported by info command. Properly log the size of files copied during the backup, matching the backup size returned from the info command. In the reference issue, the incremental backup after switchover logs the size of all files evaluated rather than only the size of the files copied in the backup.
2021-11-09 13:24:56 -05:00
TEST_RESULT_VOID(
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
lockAcquire(TEST_PATH_STR, cfgOptionStr(cfgOptStanza), cfgOptionStr(cfgOptExecId), lockTypeBackup, 0, true),
"acquire backup lock");
TEST_RESULT_VOID(
backupJobResult(manifest, STRDEF("host"), storageTest, strLstNew(), job, false, 0, &sizeProgress,
&currentPercentComplete), "log noop result");
TEST_RESULT_VOID(lockRelease(true), "release backup lock");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Increase precision of percent complete logging for backup and restore. For very large backups only getting an update per percent may not be often enough. Add hundredths to the percent complete logging to provide more timely information.
2022-03-06 14:01:24 -05:00
TEST_RESULT_LOG("P00 DETAIL: match file from prior backup host:" TEST_PATH "/test (0B, 100.00%)");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// Offline tests should only be used to test offline functionality and errors easily tested in offline mode
// *****************************************************************************************************************************
if (testBegin("cmdBackup() offline"))
{
// Set log level to detail
harnessLogLevelSet(logLevelDetail);
// Replace backup labels since the times are not deterministic
hrnLogReplaceAdd("[0-9]{8}-[0-9]{6}F_[0-9]{8}-[0-9]{6}I", NULL, "INCR", true);
hrnLogReplaceAdd("[0-9]{8}-[0-9]{6}F_[0-9]{8}-[0-9]{6}D", NULL, "DIFF", true);
hrnLogReplaceAdd("[0-9]{8}-[0-9]{6}F", NULL, "FULL", true);
// Create stanza
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
hrnCfgArgRawBool(argList, cfgOptOnline, false);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdStanzaCreate, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
// Create pg_control
Remove support for PostgreSQL 8.3/8.4. There is no evidence that users need 8.3/8.4 anymore but it does cost us in terms of development and testing, especially now that we have a number of new backup/restore features planned. It seems to make sense to remove this support now. If there are users who need to use/migrate from these versions they can use an older version of pgBackRest.
2022-01-06 15:34:04 -05:00
HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_90);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
cmdStanzaCreate();
Replace hrnLogResult() with TEST_RESULT_LOG/_FMT(). The macros provide more information when there is an error and may be updated in the future without changing the test code.
2021-05-22 14:09:45 -04:00
TEST_RESULT_LOG("P00 INFO: stanza-create for stanza 'test1' on repo1");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
Use PostgreSQL instead of postmaster where appropriate. Using postmaster in messages was not very helpful since users rarely interact directly with the postmaster. Using PostgreSQL instead seems clearer.
2020-06-17 15:14:59 -04:00
TEST_TITLE("error when pg appears to be running");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptOnline, false);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Rename PostgreSQL pid file constants and tests.
2021-10-13 19:36:59 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), PG_FILE_POSTMTRPID, "PID");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_ERROR(
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
testCmdBackup(), PgRunningError,
Use the PG_FILE_POSTMTRPID constant where appropriate. Do the same in Perl with the MANIFEST_FILE_POSTMTRPID constant.
2022-01-20 08:41:05 -05:00
"--no-online passed but " PG_FILE_POSTMTRPID " exists - looks like " PG_NAME " is running. Shut down " PG_NAME " and"
" try again, or use --force.");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_LOG("P00 WARN: no prior backup exists, incr backup has been changed to full");
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("offline full backup");
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptOnline, false);
hrnCfgArgRawBool(argList, cfgOptCompress, false);
hrnCfgArgRawBool(argList, cfgOptForce, true);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), "postgresql.conf", "CONFIGSTUFF");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_LOG_FMT(
"P00 WARN: no prior backup exists, incr backup has been changed to full\n"
Use the PG_FILE_POSTMTRPID constant where appropriate. Do the same in Perl with the MANIFEST_FILE_POSTMTRPID constant.
2022-01-20 08:41:05 -05:00
"P00 WARN: --no-online passed and " PG_FILE_POSTMTRPID " exists but --force was passed so backup will continue though"
" it looks like " PG_NAME " is running and the backup will probably not be consistent\n"
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/global/pg_control (8KB, 99.86%%) checksum %s\n"
Increase precision of percent complete logging for backup and restore. For very large backups only getting an update per percent may not be often enough. Add hundredths to the percent complete logging to provide more timely information.
2022-03-06 14:01:24 -05:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/postgresql.conf (11B, 100.00%%) checksum"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
" e3db315c260e79211b7b52587123b7aa060f30ab\n"
Log backup file total and restore size/file total. The backup size was a bit off because it did not include any files (e.g. backup_label, WAL files) that were added to the manifest after the main copy. To fix this move the log message to the very end of the backup. Add size/file total log message to restore since it did not exist before.
2021-08-11 13:39:36 -04:00
"P00 INFO: new backup label = [FULL-1]\n"
"P00 INFO: full backup size = 8KB, file total = 2",
Add support for testing on 64-bit big-endian architectures. In particular add support for s390x but we hope this will work for other 64-bit big-endian architectures. Run basic unit tests on Travis CI for 390x.
2020-07-20 09:59:16 -04:00
TEST_64BIT() ?
Add limited CI for ppc64le/s390x using emulation. The emulation is so slow that running all the unit tests would be too expensive, but this at least shows that the build works and some of the more complex tests run. In particular, it is good to test on one big-endian architecture to be sure that checksums are correct. Update checksums in the tests where they had gotten out of date since the last time we were testing on s390x. Also use a different test in command/archivePushTest to show the name of the file when a checksum does not match to aid in debugging. The command/archive-push test was updated but not included because there is also a permissions issue, which looks to be the same as what we see on MacOS/FreedBSD. Hopefully we'll be able to fix all of those at the same time.
2022-05-09 12:48:19 -04:00
(TEST_BIG_ENDIAN() ? "ec84602c8b4f62bd0ef10bd3dfcb04c3b3ce4a35" : "b7ec43e4646f5d06c95881df0c572630a1221377") :
Remove support for PostgreSQL 8.3/8.4. There is no evidence that users need 8.3/8.4 anymore but it does cost us in terms of development and testing, especially now that we have a number of new backup/restore features planned. It seems to make sense to remove this support now. If there are users who need to use/migrate from these versions they can use an older version of pgBackRest.
2022-01-06 15:34:04 -05:00
"f21ff9abdcd1ec2f600d4ee8e5792c9b61eb2e37");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Use PostgreSQL instead of postmaster where appropriate. Using postmaster in messages was not very helpful since users rarely interact directly with the postmaster. Using PostgreSQL instead seems clearer.
2020-06-17 15:14:59 -04:00
// Make pg no longer appear to be running
Rename PostgreSQL pid file constants and tests.
2021-10-13 19:36:59 -04:00
HRN_STORAGE_REMOVE(storagePgWrite(), PG_FILE_POSTMTRPID, .errorOnMissing = true);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("error when no files have changed");
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptOnline, false);
hrnCfgArgRawBool(argList, cfgOptCompress, true);
Convert BackupType enum to StringId. Allows removal of backupType()/backupTypeStr() and improves debug logging of the enum. Move BackupType enum and string constants to info/infoBackup.h so they are available to more modules. Also convert InfoBackup to use BackupType instead of a String.
2021-05-03 12:15:39 -04:00
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeDiff);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_ERROR(testCmdBackup(), FileMissingError, "no files have changed since the last backup - this seems unlikely");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_LOG(
"P00 INFO: last backup label = [FULL-1], version = " PROJECT_VERSION "\n"
Allow repo-hardlink option to be changed after full backup. This rule was added because there were not sufficient tests to demonstrate that the repo-hardlink option could be changed in a backup set. Remove the restriction and add/update tests to show that it works. This is necessary now because bundling requires that hardlinking be disabled. Rather than add code complexity, it seems better just to address this limitation.
2022-03-22 08:35:34 -06:00
"P00 WARN: diff backup cannot alter compress-type option to 'gz', reset to value in [FULL-1]");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("offline incr backup to test unresumable backup");
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptOnline, false);
hrnCfgArgRawBool(argList, cfgOptCompress, false);
hrnCfgArgRawBool(argList, cfgOptChecksumPage, true);
Convert BackupType enum to StringId. Allows removal of backupType()/backupTypeStr() and improves debug logging of the enum. Move BackupType enum and string constants to info/infoBackup.h so they are available to more modules. Also convert InfoBackup to use BackupType instead of a String.
2021-05-03 12:15:39 -04:00
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeIncr);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), PG_FILE_PGVERSION, "VER");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_LOG(
"P00 INFO: last backup label = [FULL-1], version = " PROJECT_VERSION "\n"
"P00 WARN: incr backup cannot alter 'checksum-page' option to 'true', reset to 'false' from [FULL-1]\n"
"P00 WARN: backup '[DIFF-1]' cannot be resumed: new backup type 'incr' does not match resumable backup type 'diff'\n"
Increase precision of percent complete logging for backup and restore. For very large backups only getting an update per percent may not be often enough. Add hundredths to the percent complete logging to provide more timely information.
2022-03-06 14:01:24 -05:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/PG_VERSION (3B, 100.00%) checksum c8663c2525f44b6d9c687fbceb4aafc63ed8b451\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 DETAIL: reference pg_data/global/pg_control to [FULL-1]\n"
"P00 DETAIL: reference pg_data/postgresql.conf to [FULL-1]\n"
Log backup file total and restore size/file total. The backup size was a bit off because it did not include any files (e.g. backup_label, WAL files) that were added to the manifest after the main copy. To fix this move the log message to the very end of the backup. Add size/file total log message to restore since it did not exist before.
2021-08-11 13:39:36 -04:00
"P00 INFO: new backup label = [INCR-1]\n"
"P00 INFO: incr backup size = 3B, file total = 3");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("offline diff backup to test prior backup must be full");
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRawZ(argList, cfgOptRepoPath, TEST_PATH "/repo");
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptOnline, false);
hrnCfgArgRawBool(argList, cfgOptCompress, false);
Convert BackupType enum to StringId. Allows removal of backupType()/backupTypeStr() and improves debug logging of the enum. Move BackupType enum and string constants to info/infoBackup.h so they are available to more modules. Also convert InfoBackup to use BackupType instead of a String.
2021-05-03 12:15:39 -04:00
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeDiff);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
sleepMSec(MSEC_PER_SEC - (timeMSec() % MSEC_PER_SEC));
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), PG_FILE_PGVERSION, "VR2");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_LOG(
"P00 INFO: last backup label = [FULL-1], version = " PROJECT_VERSION "\n"
Increase precision of percent complete logging for backup and restore. For very large backups only getting an update per percent may not be often enough. Add hundredths to the percent complete logging to provide more timely information.
2022-03-06 14:01:24 -05:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/PG_VERSION (3B, 100.00%) checksum 6f1894088c578e4f0b9888e8e8a997d93cbbc0c5\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 DETAIL: reference pg_data/global/pg_control to [FULL-1]\n"
"P00 DETAIL: reference pg_data/postgresql.conf to [FULL-1]\n"
Log backup file total and restore size/file total. The backup size was a bit off because it did not include any files (e.g. backup_label, WAL files) that were added to the manifest after the main copy. To fix this move the log message to the very end of the backup. Add size/file total log message to restore since it did not exist before.
2021-08-11 13:39:36 -04:00
"P00 INFO: new backup label = [DIFF-2]\n"
"P00 INFO: diff backup size = 3B, file total = 3");
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("only repo2 configured");
// Create stanza on a second repo
argList = strLstNew();
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
hrnCfgArgKeyRawZ(argList, cfgOptRepoPath, 2, TEST_PATH "/repo2");
Update CipherType/CipherMode to StringId. As in 6cc521b, this allows option values and enums to be easily mapped together.
2021-04-28 11:36:20 -04:00
hrnCfgArgKeyRawStrId(argList, cfgOptRepoCipherType, 2, cipherTypeAes256Cbc);
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
hrnCfgEnvKeyRawZ(cfgOptRepoCipherPass, 2, TEST_CIPHER_PASS);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptPgPath, TEST_PATH "/pg1");
hrnCfgArgRawBool(argList, cfgOptOnline, false);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdStanzaCreate, argList);
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
cmdStanzaCreate();
Replace hrnLogResult() with TEST_RESULT_LOG/_FMT(). The macros provide more information when there is an error and may be updated in the future without changing the test code.
2021-05-22 14:09:45 -04:00
TEST_RESULT_LOG("P00 INFO: stanza-create for stanza 'test1' on repo2");
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
// Set log level to warn
harnessLogLevelSet(logLevelWarn);
// With repo2 the only repo configured, ensure it is chosen by confirming diff is changed to full due to no prior backups
hrnCfgArgKeyRawZ(argList, cfgOptRepoRetentionFull, 2, "1");
Convert BackupType enum to StringId. Allows removal of backupType()/backupTypeStr() and improves debug logging of the enum. Move BackupType enum and string constants to info/infoBackup.h so they are available to more modules. Also convert InfoBackup to use BackupType instead of a String.
2021-05-03 12:15:39 -04:00
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeDiff);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
TEST_RESULT_LOG("P00 WARN: no prior backup exists, diff backup has been changed to full");
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("multi-repo");
Increase harness log level to detail in command/backup test. Also clean up some unneeded calls to harnessLogLevelReset().
2021-07-08 11:17:13 -04:00
// Set log level to detail
harnessLogLevelSet(logLevelDetail);
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
// Add repo1 to the configuration
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgKeyRawZ(argList, cfgOptRepoPath, 1, TEST_PATH "/repo");
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
hrnCfgArgKeyRawZ(argList, cfgOptRepoRetentionFull, 1, "1");
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
TEST_RESULT_LOG(
"P00 INFO: repo option not specified, defaulting to repo1\n"
"P00 INFO: last backup label = [FULL-1], version = " PROJECT_VERSION "\n"
"P00 WARN: diff backup cannot alter compress-type option to 'gz', reset to value in [FULL-1]\n"
Increase precision of percent complete logging for backup and restore. For very large backups only getting an update per percent may not be often enough. Add hundredths to the percent complete logging to provide more timely information.
2022-03-06 14:01:24 -05:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/PG_VERSION (3B, 100.00%) checksum 6f1894088c578e4f0b9888e8e8a997d93cbbc0c5\n"
Increase harness log level to detail in command/backup test. Also clean up some unneeded calls to harnessLogLevelReset().
2021-07-08 11:17:13 -04:00
"P00 DETAIL: reference pg_data/global/pg_control to [FULL-1]\n"
"P00 DETAIL: reference pg_data/postgresql.conf to [FULL-1]\n"
Log backup file total and restore size/file total. The backup size was a bit off because it did not include any files (e.g. backup_label, WAL files) that were added to the manifest after the main copy. To fix this move the log message to the very end of the backup. Add size/file total log message to restore since it did not exist before.
2021-08-11 13:39:36 -04:00
"P00 INFO: new backup label = [DIFF-3]\n"
"P00 INFO: diff backup size = 3B, file total = 3");
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("multi-repo - specify repo");
hrnCfgArgRawZ(argList, cfgOptRepo, "2");
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), PG_FILE_PGVERSION, "VER");
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
unsigned int backupCount = strLstSize(storageListP(storageRepoIdx(1), strNewFmt(STORAGE_PATH_BACKUP "/test1")));
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
TEST_RESULT_LOG(
"P00 INFO: last backup label = [FULL-2], version = " PROJECT_VERSION "\n"
Increase precision of percent complete logging for backup and restore. For very large backups only getting an update per percent may not be often enough. Add hundredths to the percent complete logging to provide more timely information.
2022-03-06 14:01:24 -05:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/PG_VERSION (3B, 100.00%) checksum c8663c2525f44b6d9c687fbceb4aafc63ed8b451\n"
Increase harness log level to detail in command/backup test. Also clean up some unneeded calls to harnessLogLevelReset().
2021-07-08 11:17:13 -04:00
"P00 DETAIL: reference pg_data/global/pg_control to [FULL-2]\n"
"P00 DETAIL: reference pg_data/postgresql.conf to [FULL-2]\n"
Log backup file total and restore size/file total. The backup size was a bit off because it did not include any files (e.g. backup_label, WAL files) that were added to the manifest after the main copy. To fix this move the log message to the very end of the backup. Add size/file total log message to restore since it did not exist before.
2021-08-11 13:39:36 -04:00
"P00 INFO: new backup label = [DIFF-4]\n"
"P00 INFO: diff backup size = 3B, file total = 3");
Make --repo optional for backup command. If there are multiple repos and the --repo option is not specified then backup will automatically select the highest priority repo.
2021-02-26 14:49:50 -05:00
TEST_RESULT_UINT(
strLstSize(storageListP(storageRepoIdx(1), strNewFmt(STORAGE_PATH_BACKUP "/test1"))), backupCount + 1,
"new backup repo2");
// Cleanup
hrnCfgEnvKeyRemoveRaw(cfgOptRepoCipherPass, 2);
harnessLogLevelReset();
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// *****************************************************************************************************************************
if (testBegin("cmdBackup() online"))
{
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
const String *pg1Path = STRDEF(TEST_PATH "/pg1");
const String *repoPath = STRDEF(TEST_PATH "/repo");
const String *pg2Path = STRDEF(TEST_PATH "/pg2");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Set log level to detail
harnessLogLevelSet(logLevelDetail);
// Replace percent complete and backup size since they can cause a lot of churn when files are added/removed
Increase precision of percent complete logging for backup and restore. For very large backups only getting an update per percent may not be often enough. Add hundredths to the percent complete logging to provide more timely information.
2022-03-06 14:01:24 -05:00
hrnLogReplaceAdd(", [0-9]{1,3}.[0-9]{1,2}%\\)", "[0-9].+%", "PCT", false);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
hrnLogReplaceAdd(" backup size = [0-9]+[A-Z]+", "[^ ]+$", "SIZE", false);
// Replace checksums since they can differ between architectures (e.g. 32/64 bit)
hrnLogReplaceAdd("\\) checksum [a-f0-9]{40}", "[a-f0-9]{40}$", "SHA1", false);
// Backup start time epoch. The idea is to not have backup times (and therefore labels) ever change. Each backup added
// should be separated by 100,000 seconds (1,000,000 after stanza-upgrade) but after the initial assignments this will only
// be possible at the beginning and the end, so new backups added in the middle will average the start times of the prior
// and next backup to get their start time. Backups added to the beginning of the test will need to subtract from the
// epoch.
#define BACKUP_EPOCH 1570000000
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("online 9.5 resume uncompressed full backup");
time_t backupTimeStart = BACKUP_EPOCH;
{
// Create stanza
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRaw(argList, cfgOptRepoPath, repoPath);
hrnCfgArgRaw(argList, cfgOptPgPath, pg1Path);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptOnline, false);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdStanzaCreate, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
// Create pg_control
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_95);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
cmdStanzaCreate();
Replace hrnLogResult() with TEST_RESULT_LOG/_FMT(). The macros provide more information when there is an error and may be updated in the future without changing the test code.
2021-05-22 14:09:45 -04:00
TEST_RESULT_LOG("P00 INFO: stanza-create for stanza 'test1' on repo1");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Load options
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRaw(argList, cfgOptRepoPath, repoPath);
hrnCfgArgRaw(argList, cfgOptPgPath, pg1Path);
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Convert BackupType enum to StringId. Allows removal of backupType()/backupTypeStr() and improves debug logging of the enum. Move BackupType enum and string constants to info/infoBackup.h so they are available to more modules. Also convert InfoBackup to use BackupType instead of a String.
2021-05-03 12:15:39 -04:00
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeFull);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptStopAuto, true);
hrnCfgArgRawBool(argList, cfgOptCompress, false);
hrnCfgArgRawBool(argList, cfgOptArchiveCheck, false);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Add files
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), "postgresql.conf", "CONFIGSTUFF", .timeModified = backupTimeStart);
HRN_STORAGE_PUT_Z(storagePgWrite(), PG_FILE_PGVERSION, PG_VERSION_95_STR, .timeModified = backupTimeStart);
HRN_STORAGE_PATH_CREATE(storagePgWrite(), strZ(pgWalPath(PG_VERSION_95)), .noParentCreate = true);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Create a backup manifest that looks like a halted backup manifest
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
Manifest *manifestResume = manifestNewBuild(
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
storagePg(), PG_VERSION_95, hrnPgCatalogVersion(PG_VERSION_95), true, false, false, NULL, NULL);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
ManifestData *manifestResumeData = (ManifestData *)manifestData(manifestResume);
manifestResumeData->backupType = backupTypeFull;
const String *resumeLabel = backupLabelCreate(backupTypeFull, NULL, backupTimeStart);
manifestBackupLabelSet(manifestResume, resumeLabel);
// Copy a file to be resumed that has not changed in the repo
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_COPY(
storagePg(), PG_FILE_PGVERSION, storageRepoWrite(),
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/PG_VERSION", strZ(resumeLabel)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Pack manifest file structs to save memory. Manifests with a very large number of files can use a considerable amount of memory. There are a lot of zeroes in the data so it can be stored more efficiently by using base-128 varint encoding for the integers and storing the strings in the same allocation. The downside is that the data needs to be unpacked in order to be used, but in most cases this seems fast enough (about 10% slower than before) except for saving the manifest, which is 10% slower up to 10 million files and then gets about 5x slower by 100 million (two minutes on my M1 Mac). Profiling does not show this slowdown so I wonder if this is related to the change in memory layout. Curiously, the function that increased most was jsonFromStrInternal(), which was not modified. That gives more weight to the idea that there is some kind of memory issue going on here and one hopes that servers would be less affected. Either way, they largest use cases we have seen are for about 6 million files so if we can improve that case I believe we will be better off. Further analysis showed that most of the time was taken up writing the size and timestamp fields, which makes almost no sense. The same amount of time was used if they were hard-coded to 0, which points to some odd memory issue on the M1 architecture. This change has been planned for a while, but the particular impetus at this time is that small file support requires additional fields that would increase manifest memory usage by about 20%, even if the feature is not used. Note that the Pack code has been updated to use the new varint encoder, but the decoder remains separate because it needs to fetch one byte at a time.
2022-01-21 17:05:07 -05:00
ManifestFilePack **const filePack = manifestFilePackFindInternal(manifestResume, STRDEF("pg_data/PG_VERSION"));
Simplify manifest file defaults. Previously manifest load required two passes through the file list, one to load the data and one to set the defaults. This required each file to be packed twice. Instead simply note that the file value is default and then set the file defaults when they are loaded from the manifest. This is made possible by the different internal/external representations for files so the same method cannot be applied to paths and links. This change seems to resolve the performance issues noted in 61ce586 but there is no obvious reason why.
2022-01-24 16:21:07 -05:00
ManifestFile file = manifestFileUnpack(manifestResume, *filePack);
Pack manifest file structs to save memory. Manifests with a very large number of files can use a considerable amount of memory. There are a lot of zeroes in the data so it can be stored more efficiently by using base-128 varint encoding for the integers and storing the strings in the same allocation. The downside is that the data needs to be unpacked in order to be used, but in most cases this seems fast enough (about 10% slower than before) except for saving the manifest, which is 10% slower up to 10 million files and then gets about 5x slower by 100 million (two minutes on my M1 Mac). Profiling does not show this slowdown so I wonder if this is related to the change in memory layout. Curiously, the function that increased most was jsonFromStrInternal(), which was not modified. That gives more weight to the idea that there is some kind of memory issue going on here and one hopes that servers would be less affected. Either way, they largest use cases we have seen are for about 6 million files so if we can improve that case I believe we will be better off. Further analysis showed that most of the time was taken up writing the size and timestamp fields, which makes almost no sense. The same amount of time was used if they were hard-coded to 0, which points to some odd memory issue on the M1 architecture. This change has been planned for a while, but the particular impetus at this time is that small file support requires additional fields that would increase manifest memory usage by about 20%, even if the feature is not used. Note that the Pack code has been updated to use the new varint encoder, but the decoder remains separate because it needs to fetch one byte at a time.
2022-01-21 17:05:07 -05:00
strcpy(file.checksumSha1, "06d06bb31b570b94d7b4325f511f853dbe771c21");
manifestFilePackUpdate(manifestResume, filePack, &file);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Save the resume manifest
manifestSave(
manifestResume,
storageWriteIo(
storageNewWriteP(
storageRepoWrite(),
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
strNewFmt(STORAGE_REPO_BACKUP "/%s/" BACKUP_MANIFEST_FILE INFO_COPY_EXT, strZ(resumeLabel)))));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Run backup
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
testBackupPqScriptP(PG_VERSION_95, backupTimeStart, .noArchiveCheck = true, .noWal = true);
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_LOG(
"P00 INFO: execute exclusive pg_start_backup(): backup begins after the next regular checkpoint completes\n"
"P00 INFO: backup start archive = 0000000105D944C000000000, lsn = 5d944c0/0\n"
"P00 WARN: resumable backup 20191002-070640F of same type exists -- remove invalid files and resume\n"
Change level of backup/restore copied file logging to detail. The log level for copied files in the backup/restore commands has been changed to detail. This makes the info log level less noisy but if these messages are required then set the log level for the backup/restore commands to detail.
2021-07-09 13:50:35 -04:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/global/pg_control (8KB, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/postgresql.conf (11B, [PCT]) checksum [SHA1]\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"P01 DETAIL: checksum resumed file " TEST_PATH "/pg1/PG_VERSION (3B, [PCT]) checksum [SHA1]\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 INFO: execute exclusive pg_stop_backup() and wait for all WAL segments to archive\n"
"P00 INFO: backup stop archive = 0000000105D944C000000000, lsn = 5d944c0/800000\n"
Log backup file total and restore size/file total. The backup size was a bit off because it did not include any files (e.g. backup_label, WAL files) that were added to the manifest after the main copy. To fix this move the log message to the very end of the backup. Add size/file total log message to restore since it did not exist before.
2021-08-11 13:39:36 -04:00
"P00 INFO: new backup label = 20191002-070640F\n"
"P00 INFO: full backup size = [SIZE], file total = 3");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Remove last vestiges of key replacement missed in b270253a.
2021-07-15 13:19:49 -04:00
TEST_RESULT_STR_Z(
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
testBackupValidate(storageRepo(), STRDEF(STORAGE_REPO_BACKUP "/latest")),
". {link, d=20191002-070640F}\n"
"pg_data {path}\n"
"pg_data/PG_VERSION {file, s=3}\n"
"pg_data/global {path}\n"
"pg_data/global/pg_control {file, s=8192}\n"
"pg_data/pg_xlog {path}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"pg_data/postgresql.conf {file, s=11}\n"
"--------\n"
"[backup:target]\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"pg_data={\"path\":\"" TEST_PATH "/pg1\",\"type\":\"path\"}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"\n"
"[target:file]\n"
"pg_data/PG_VERSION={\"checksum\":\"06d06bb31b570b94d7b4325f511f853dbe771c21\",\"size\":3"
",\"timestamp\":1570000000}\n"
"pg_data/global/pg_control={\"size\":8192,\"timestamp\":1570000000}\n"
"pg_data/postgresql.conf={\"checksum\":\"e3db315c260e79211b7b52587123b7aa060f30ab\",\"size\":11"
",\"timestamp\":1570000000}\n"
"\n"
"[target:path]\n"
"pg_data={}\n"
"pg_data/global={}\n"
"pg_data/pg_xlog={}\n",
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"compare file list");
}
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("online resumed compressed 9.5 full backup");
// Backup start time
backupTimeStart = BACKUP_EPOCH + 100000;
{
// Load options
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRaw(argList, cfgOptRepoPath, repoPath);
hrnCfgArgRaw(argList, cfgOptPgPath, pg1Path);
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Convert BackupType enum to StringId. Allows removal of backupType()/backupTypeStr() and improves debug logging of the enum. Move BackupType enum and string constants to info/infoBackup.h so they are available to more modules. Also convert InfoBackup to use BackupType instead of a String.
2021-05-03 12:15:39 -04:00
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeFull);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptStopAuto, true);
hrnCfgArgRawBool(argList, cfgOptArchiveCopy, true);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Create a backup manifest that looks like a halted backup manifest
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
Manifest *manifestResume = manifestNewBuild(
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
storagePg(), PG_VERSION_95, hrnPgCatalogVersion(PG_VERSION_95), true, false, false, NULL, NULL);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
ManifestData *manifestResumeData = (ManifestData *)manifestData(manifestResume);
manifestResumeData->backupType = backupTypeFull;
Add infrastructure for multiple compression type support. Add compress-type option and deprecate compress option. Since the compress option is boolean it won't work with multiple compression types. Add logic to cfgLoadUpdateOption() to update compress-type if it is not set directly. The compress option should no longer be referenced outside the cfgLoadUpdateOption() function. Add common/compress/helper module to contain interface functions that work with multiple compression types. Code outside this module should no longer call specific compression drivers, though it may be OK to reference a specific compression type using the new interface (e.g., saving backup history files in gz format). Unit tests only test compression using the gz format because other formats may not be available in all builds. It is the job of integration tests to exercise all compression types. Additional compression types will be added in future commits.
2020-03-06 14:41:03 -05:00
manifestResumeData->backupOptionCompressType = compressTypeGz;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
const String *resumeLabel = backupLabelCreate(backupTypeFull, NULL, backupTimeStart);
manifestBackupLabelSet(manifestResume, resumeLabel);
// File exists in cluster and repo but not in the resume manifest
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), "not-in-resume", "TEST", .timeModified = backupTimeStart);
HRN_STORAGE_PUT_EMPTY(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/not-in-resume.gz", strZ(resumeLabel)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Remove checksum from file so it won't be resumed
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_EMPTY(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/global/pg_control.gz", strZ(resumeLabel)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Pack manifest file structs to save memory. Manifests with a very large number of files can use a considerable amount of memory. There are a lot of zeroes in the data so it can be stored more efficiently by using base-128 varint encoding for the integers and storing the strings in the same allocation. The downside is that the data needs to be unpacked in order to be used, but in most cases this seems fast enough (about 10% slower than before) except for saving the manifest, which is 10% slower up to 10 million files and then gets about 5x slower by 100 million (two minutes on my M1 Mac). Profiling does not show this slowdown so I wonder if this is related to the change in memory layout. Curiously, the function that increased most was jsonFromStrInternal(), which was not modified. That gives more weight to the idea that there is some kind of memory issue going on here and one hopes that servers would be less affected. Either way, they largest use cases we have seen are for about 6 million files so if we can improve that case I believe we will be better off. Further analysis showed that most of the time was taken up writing the size and timestamp fields, which makes almost no sense. The same amount of time was used if they were hard-coded to 0, which points to some odd memory issue on the M1 architecture. This change has been planned for a while, but the particular impetus at this time is that small file support requires additional fields that would increase manifest memory usage by about 20%, even if the feature is not used. Note that the Pack code has been updated to use the new varint encoder, but the decoder remains separate because it needs to fetch one byte at a time.
2022-01-21 17:05:07 -05:00
ManifestFilePack **const filePack = manifestFilePackFindInternal(manifestResume, STRDEF("pg_data/global/pg_control"));
Simplify manifest file defaults. Previously manifest load required two passes through the file list, one to load the data and one to set the defaults. This required each file to be packed twice. Instead simply note that the file value is default and then set the file defaults when they are loaded from the manifest. This is made possible by the different internal/external representations for files so the same method cannot be applied to paths and links. This change seems to resolve the performance issues noted in 61ce586 but there is no obvious reason why.
2022-01-24 16:21:07 -05:00
ManifestFile file = manifestFileUnpack(manifestResume, *filePack);
Pack manifest file structs to save memory. Manifests with a very large number of files can use a considerable amount of memory. There are a lot of zeroes in the data so it can be stored more efficiently by using base-128 varint encoding for the integers and storing the strings in the same allocation. The downside is that the data needs to be unpacked in order to be used, but in most cases this seems fast enough (about 10% slower than before) except for saving the manifest, which is 10% slower up to 10 million files and then gets about 5x slower by 100 million (two minutes on my M1 Mac). Profiling does not show this slowdown so I wonder if this is related to the change in memory layout. Curiously, the function that increased most was jsonFromStrInternal(), which was not modified. That gives more weight to the idea that there is some kind of memory issue going on here and one hopes that servers would be less affected. Either way, they largest use cases we have seen are for about 6 million files so if we can improve that case I believe we will be better off. Further analysis showed that most of the time was taken up writing the size and timestamp fields, which makes almost no sense. The same amount of time was used if they were hard-coded to 0, which points to some odd memory issue on the M1 architecture. This change has been planned for a while, but the particular impetus at this time is that small file support requires additional fields that would increase manifest memory usage by about 20%, even if the feature is not used. Note that the Pack code has been updated to use the new varint encoder, but the decoder remains separate because it needs to fetch one byte at a time.
2022-01-21 17:05:07 -05:00
file.checksumSha1[0] = 0;
manifestFilePackUpdate(manifestResume, filePack, &file);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Size does not match between cluster and resume manifest
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), "size-mismatch", "TEST", .timeModified = backupTimeStart);
HRN_STORAGE_PUT_EMPTY(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/size-mismatch.gz", strZ(resumeLabel)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
manifestFileAdd(
Pass file by reference in manifestFileAdd(). Coverity complained that this pass by value was inefficient: CID 376402: Performance inefficiencies (PASS_BY_VALUE) Passing parameter file of type "ManifestFile" (size 136 bytes) by value. This was completely intentional since it gives us a copy of the struct that we can change without bothering the caller. However, updating fields is fine and may benefit the caller at some future data, and in any case does no harm now. And as usual it is easier not to fight with Coverity.
2022-02-20 16:45:07 -06:00
manifestResume, &(ManifestFile){
Fix spacing and typos in backup/manifest modules.
2020-11-09 16:26:43 -05:00
.name = STRDEF("pg_data/size-mismatch"), .checksumSha1 = "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
.size = 33});
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Time does not match between cluster and resume manifest
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), "time-mismatch", "TEST", .timeModified = backupTimeStart);
HRN_STORAGE_PUT_EMPTY(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/time-mismatch.gz", strZ(resumeLabel)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
manifestFileAdd(
Pass file by reference in manifestFileAdd(). Coverity complained that this pass by value was inefficient: CID 376402: Performance inefficiencies (PASS_BY_VALUE) Passing parameter file of type "ManifestFile" (size 136 bytes) by value. This was completely intentional since it gives us a copy of the struct that we can change without bothering the caller. However, updating fields is fine and may benefit the caller at some future data, and in any case does no harm now. And as usual it is easier not to fight with Coverity.
2022-02-20 16:45:07 -06:00
manifestResume, &(ManifestFile){
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
.name = STRDEF("pg_data/time-mismatch"), .checksumSha1 = "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx", .size = 4,
.timestamp = backupTimeStart - 1});
// Size is zero in cluster and resume manifest. ??? We'd like to remove this requirement after the migration.
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_EMPTY(storagePgWrite(), "zero-size", .timeModified = backupTimeStart);
HRN_STORAGE_PUT_Z(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/zero-size.gz", strZ(resumeLabel)), "ZERO-SIZE");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
manifestFileAdd(
Pass file by reference in manifestFileAdd(). Coverity complained that this pass by value was inefficient: CID 376402: Performance inefficiencies (PASS_BY_VALUE) Passing parameter file of type "ManifestFile" (size 136 bytes) by value. This was completely intentional since it gives us a copy of the struct that we can change without bothering the caller. However, updating fields is fine and may benefit the caller at some future data, and in any case does no harm now. And as usual it is easier not to fight with Coverity.
2022-02-20 16:45:07 -06:00
manifestResume, &(ManifestFile){.name = STRDEF("pg_data/zero-size"), .size = 0, .timestamp = backupTimeStart});
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Path is not in manifest
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
HRN_STORAGE_PATH_CREATE(storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/bogus_path", strZ(resumeLabel)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// File is not in manifest
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_EMPTY(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/global/bogus.gz", strZ(resumeLabel)));
Add infrastructure for multiple compression type support. Add compress-type option and deprecate compress option. Since the compress option is boolean it won't work with multiple compression types. Add logic to cfgLoadUpdateOption() to update compress-type if it is not set directly. The compress option should no longer be referenced outside the cfgLoadUpdateOption() function. Add common/compress/helper module to contain interface functions that work with multiple compression types. Code outside this module should no longer call specific compression drivers, though it may be OK to reference a specific compression type using the new interface (e.g., saving backup history files in gz format). Unit tests only test compression using the gz format because other formats may not be available in all builds. It is the job of integration tests to exercise all compression types. Additional compression types will be added in future commits.
2020-03-06 14:41:03 -05:00
// File has incorrect compression type
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
HRN_STORAGE_PUT_EMPTY(storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/global/bogus", strZ(resumeLabel)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Save the resume manifest
manifestSave(
manifestResume,
storageWriteIo(
storageNewWriteP(
storageRepoWrite(),
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
strNewFmt(STORAGE_REPO_BACKUP "/%s/" BACKUP_MANIFEST_FILE INFO_COPY_EXT, strZ(resumeLabel)))));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Disable storageFeaturePath so paths will not be created before files are copied
Refactor storage modules with inline getters/setters. Extended the pattern introduced in 79a2d02c to the storage modules: Storage, StorageRead, StorageWrite.
2021-04-07 14:04:38 -04:00
((Storage *)storageRepoWrite())->pub.interface.feature ^= 1 << storageFeaturePath;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Disable storageFeaturePathSync so paths will not be synced
Refactor storage modules with inline getters/setters. Extended the pattern introduced in 79a2d02c to the storage modules: Storage, StorageRead, StorageWrite.
2021-04-07 14:04:38 -04:00
((Storage *)storageRepoWrite())->pub.interface.feature ^= 1 << storageFeaturePathSync;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Run backup
testBackupPqScriptP(PG_VERSION_95, backupTimeStart);
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Enable storage features
Refactor storage modules with inline getters/setters. Extended the pattern introduced in 79a2d02c to the storage modules: Storage, StorageRead, StorageWrite.
2021-04-07 14:04:38 -04:00
((Storage *)storageRepoWrite())->pub.interface.feature |= 1 << storageFeaturePath;
((Storage *)storageRepoWrite())->pub.interface.feature |= 1 << storageFeaturePathSync;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_LOG(
"P00 INFO: execute exclusive pg_start_backup(): backup begins after the next regular checkpoint completes\n"
"P00 INFO: backup start archive = 0000000105D95D3000000000, lsn = 5d95d30/0\n"
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
"P00 INFO: check archive for prior segment 0000000105D95D2F000000FF\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 WARN: resumable backup 20191003-105320F of same type exists -- remove invalid files and resume\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"P00 DETAIL: remove path '" TEST_PATH "/repo/backup/test1/20191003-105320F/pg_data/bogus_path' from resumed"
" backup\n"
"P00 DETAIL: remove file '" TEST_PATH "/repo/backup/test1/20191003-105320F/pg_data/global/bogus' from resumed"
" backup (mismatched compression type)\n"
"P00 DETAIL: remove file '" TEST_PATH "/repo/backup/test1/20191003-105320F/pg_data/global/bogus.gz' from resumed"
" backup (missing in manifest)\n"
"P00 DETAIL: remove file '" TEST_PATH "/repo/backup/test1/20191003-105320F/pg_data/global/pg_control.gz' from"
" resumed backup (no checksum in resumed manifest)\n"
"P00 DETAIL: remove file '" TEST_PATH "/repo/backup/test1/20191003-105320F/pg_data/not-in-resume.gz' from resumed"
" backup (missing in resumed manifest)\n"
"P00 DETAIL: remove file '" TEST_PATH "/repo/backup/test1/20191003-105320F/pg_data/size-mismatch.gz' from resumed"
" backup (mismatched size)\n"
"P00 DETAIL: remove file '" TEST_PATH "/repo/backup/test1/20191003-105320F/pg_data/time-mismatch.gz' from resumed"
" backup (mismatched timestamp)\n"
"P00 DETAIL: remove file '" TEST_PATH "/repo/backup/test1/20191003-105320F/pg_data/zero-size.gz' from resumed"
" backup (zero size)\n"
Change level of backup/restore copied file logging to detail. The log level for copied files in the backup/restore commands has been changed to detail. This makes the info log level less noisy but if these messages are required then set the log level for the backup/restore commands to detail.
2021-07-09 13:50:35 -04:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/global/pg_control (8KB, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/postgresql.conf (11B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/time-mismatch (4B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/size-mismatch (4B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/not-in-resume (4B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/PG_VERSION (3B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/zero-size (0B, [PCT])\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 INFO: execute exclusive pg_stop_backup() and wait for all WAL segments to archive\n"
"P00 INFO: backup stop archive = 0000000105D95D3000000000, lsn = 5d95d30/800000\n"
"P00 INFO: check archive for segment(s) 0000000105D95D3000000000:0000000105D95D3000000000\n"
Log archive copy during backup. Copying can be a fairly expensive operation so it makes sense to log it so the user gets some status during long copy operations.
2021-03-11 08:22:44 -05:00
"P00 DETAIL: copy segment 0000000105D95D3000000000 to backup\n"
Log backup file total and restore size/file total. The backup size was a bit off because it did not include any files (e.g. backup_label, WAL files) that were added to the manifest after the main copy. To fix this move the log message to the very end of the backup. Add size/file total log message to restore since it did not exist before.
2021-08-11 13:39:36 -04:00
"P00 INFO: new backup label = 20191003-105320F\n"
"P00 INFO: full backup size = [SIZE], file total = 8");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Remove last vestiges of key replacement missed in b270253a.
2021-07-15 13:19:49 -04:00
TEST_RESULT_STR_Z(
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
testBackupValidate(storageRepo(), STRDEF(STORAGE_REPO_BACKUP "/latest")),
". {link, d=20191003-105320F}\n"
"pg_data {path}\n"
"pg_data/PG_VERSION.gz {file, s=3}\n"
"pg_data/global {path}\n"
"pg_data/global/pg_control.gz {file, s=8192}\n"
"pg_data/not-in-resume.gz {file, s=4}\n"
"pg_data/pg_xlog {path}\n"
"pg_data/pg_xlog/0000000105D95D3000000000.gz {file, s=16777216}\n"
"pg_data/postgresql.conf.gz {file, s=11}\n"
"pg_data/size-mismatch.gz {file, s=4}\n"
"pg_data/time-mismatch.gz {file, s=4}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"pg_data/zero-size.gz {file, s=0}\n"
"--------\n"
"[backup:target]\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"pg_data={\"path\":\"" TEST_PATH "/pg1\",\"type\":\"path\"}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"\n"
"[target:file]\n"
"pg_data/PG_VERSION={\"checksum\":\"06d06bb31b570b94d7b4325f511f853dbe771c21\",\"size\":3"
",\"timestamp\":1570000000}\n"
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
"pg_data/global/pg_control={\"size\":8192,\"timestamp\":1570100000}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"pg_data/not-in-resume={\"checksum\":\"984816fd329622876e14907634264e6f332e9fb3\",\"size\":4"
",\"timestamp\":1570100000}\n"
"pg_data/pg_xlog/0000000105D95D3000000000={\"size\":16777216,\"timestamp\":1570100002}\n"
"pg_data/postgresql.conf={\"checksum\":\"e3db315c260e79211b7b52587123b7aa060f30ab\",\"size\":11"
",\"timestamp\":1570000000}\n"
"pg_data/size-mismatch={\"checksum\":\"984816fd329622876e14907634264e6f332e9fb3\",\"size\":4"
",\"timestamp\":1570100000}\n"
"pg_data/time-mismatch={\"checksum\":\"984816fd329622876e14907634264e6f332e9fb3\",\"size\":4"
",\"timestamp\":1570100000}\n"
"pg_data/zero-size={\"size\":0,\"timestamp\":1570100000}\n"
"\n"
"[target:path]\n"
"pg_data={}\n"
"pg_data/global={}\n"
"pg_data/pg_xlog={}\n",
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"compare file list");
// Remove test files
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_REMOVE(storagePgWrite(), "not-in-resume", .errorOnMissing = true);
HRN_STORAGE_REMOVE(storagePgWrite(), "size-mismatch", .errorOnMissing = true);
HRN_STORAGE_REMOVE(storagePgWrite(), "time-mismatch", .errorOnMissing = true);
HRN_STORAGE_REMOVE(storagePgWrite(), "zero-size", .errorOnMissing = true);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("online resumed compressed 9.5 diff backup");
backupTimeStart = BACKUP_EPOCH + 200000;
{
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRaw(argList, cfgOptRepoPath, repoPath);
hrnCfgArgRaw(argList, cfgOptPgPath, pg1Path);
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Convert BackupType enum to StringId. Allows removal of backupType()/backupTypeStr() and improves debug logging of the enum. Move BackupType enum and string constants to info/infoBackup.h so they are available to more modules. Also convert InfoBackup to use BackupType instead of a String.
2021-05-03 12:15:39 -04:00
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeDiff);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptCompress, false);
hrnCfgArgRawBool(argList, cfgOptStopAuto, true);
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawBool(argList, cfgOptRepoHardlink, true);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Load the previous manifest and null out the checksum-page option to be sure it gets set to false in this backup
const String *manifestPriorFile = STRDEF(STORAGE_REPO_BACKUP "/latest/" BACKUP_MANIFEST_FILE);
Manifest *manifestPrior = manifestNewLoad(storageReadIo(storageNewReadP(storageRepo(), manifestPriorFile)));
((ManifestData *)manifestData(manifestPrior))->backupOptionChecksumPage = NULL;
manifestSave(manifestPrior, storageWriteIo(storageNewWriteP(storageRepoWrite(), manifestPriorFile)));
// Create a backup manifest that looks like a halted backup manifest
Improve PostgreSQL version identification. Previously, catalog versions were fixed for all versions which made maintaining the catalog versions during PostgreSQL beta and release candidate cycles very painful. A version of pgBackRest which was functionally compatible was rendered useless by a catalog version bump in PostgreSQL. Instead use only the control version to identify a PostgreSQL version when possible. Some older versions require a catalog version to positively identify a PostgreSQL version, so include them when required. Since the catalog number is required to work with tablespaces it will need to be stored. There's already a copy of it in backup.info so use that (even though we have been ignoring it in the C versions).
2020-09-18 16:55:26 -04:00
Manifest *manifestResume = manifestNewBuild(
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
storagePg(), PG_VERSION_95, hrnPgCatalogVersion(PG_VERSION_95), true, false, false, NULL, NULL);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
ManifestData *manifestResumeData = (ManifestData *)manifestData(manifestResume);
manifestResumeData->backupType = backupTypeDiff;
manifestResumeData->backupLabelPrior = manifestData(manifestPrior)->backupLabel;
Add infrastructure for multiple compression type support. Add compress-type option and deprecate compress option. Since the compress option is boolean it won't work with multiple compression types. Add logic to cfgLoadUpdateOption() to update compress-type if it is not set directly. The compress option should no longer be referenced outside the cfgLoadUpdateOption() function. Add common/compress/helper module to contain interface functions that work with multiple compression types. Code outside this module should no longer call specific compression drivers, though it may be OK to reference a specific compression type using the new interface (e.g., saving backup history files in gz format). Unit tests only test compression using the gz format because other formats may not be available in all builds. It is the job of integration tests to exercise all compression types. Additional compression types will be added in future commits.
2020-03-06 14:41:03 -05:00
manifestResumeData->backupOptionCompressType = compressTypeGz;
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
const String *resumeLabel = backupLabelCreate(
backupTypeDiff, manifestData(manifestPrior)->backupLabel, backupTimeStart);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
manifestBackupLabelSet(manifestResume, resumeLabel);
// Reference in manifest
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
HRN_STORAGE_PUT_EMPTY(storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/PG_VERSION.gz", strZ(resumeLabel)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Reference in resumed manifest
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_EMPTY(storagePgWrite(), "resume-ref", .timeModified = backupTimeStart);
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
HRN_STORAGE_PUT_EMPTY(storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/resume-ref.gz", strZ(resumeLabel)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
manifestFileAdd(
Pass file by reference in manifestFileAdd(). Coverity complained that this pass by value was inefficient: CID 376402: Performance inefficiencies (PASS_BY_VALUE) Passing parameter file of type "ManifestFile" (size 136 bytes) by value. This was completely intentional since it gives us a copy of the struct that we can change without bothering the caller. However, updating fields is fine and may benefit the caller at some future data, and in any case does no harm now. And as usual it is easier not to fight with Coverity.
2022-02-20 16:45:07 -06:00
manifestResume, &(ManifestFile){.name = STRDEF("pg_data/resume-ref"), .size = 0, .reference = STRDEF("BOGUS")});
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Fix spacing and typos in backup/manifest modules.
2020-11-09 16:26:43 -05:00
// Time does not match between cluster and resume manifest (but resume because time is in future so delta enabled). Note
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// also that the repo file is intenionally corrupt to generate a warning about corruption in the repository.
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), "time-mismatch2", "TEST", .timeModified = backupTimeStart + 100);
HRN_STORAGE_PUT_EMPTY(
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
storageRepoWrite(), zNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/time-mismatch2.gz", strZ(resumeLabel)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
manifestFileAdd(
Pass file by reference in manifestFileAdd(). Coverity complained that this pass by value was inefficient: CID 376402: Performance inefficiencies (PASS_BY_VALUE) Passing parameter file of type "ManifestFile" (size 136 bytes) by value. This was completely intentional since it gives us a copy of the struct that we can change without bothering the caller. However, updating fields is fine and may benefit the caller at some future data, and in any case does no harm now. And as usual it is easier not to fight with Coverity.
2022-02-20 16:45:07 -06:00
manifestResume, &(ManifestFile){
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
.name = STRDEF("pg_data/time-mismatch2"), .checksumSha1 = "984816fd329622876e14907634264e6f332e9fb3", .size = 4,
.timestamp = backupTimeStart});
// Links are always removed on resume
THROW_ON_SYS_ERROR(
symlink(
"..",
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
strZ(storagePathP(storageRepo(), strNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/link", strZ(resumeLabel))))) == -1,
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
FileOpenError, "unable to create symlink");
// Special files should not be in the repo
Rename TEST_SYSTEM*() to HRN_SYSTEM*(). These calls are not tests, rather they setup data for tests.
2021-05-22 14:22:51 -04:00
HRN_SYSTEM_FMT(
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"mkfifo -m 666 %s",
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
strZ(storagePathP(storageRepo(), strNewFmt(STORAGE_REPO_BACKUP "/%s/pg_data/pipe", strZ(resumeLabel)))));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Save the resume manifest
manifestSave(
manifestResume,
storageWriteIo(
storageNewWriteP(
storageRepoWrite(),
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
strNewFmt(STORAGE_REPO_BACKUP "/%s/" BACKUP_MANIFEST_FILE INFO_COPY_EXT, strZ(resumeLabel)))));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Run backup
testBackupPqScriptP(PG_VERSION_95, backupTimeStart);
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Check log
TEST_RESULT_LOG(
"P00 INFO: last backup label = 20191003-105320F, version = " PROJECT_VERSION "\n"
Add infrastructure for multiple compression type support. Add compress-type option and deprecate compress option. Since the compress option is boolean it won't work with multiple compression types. Add logic to cfgLoadUpdateOption() to update compress-type if it is not set directly. The compress option should no longer be referenced outside the cfgLoadUpdateOption() function. Add common/compress/helper module to contain interface functions that work with multiple compression types. Code outside this module should no longer call specific compression drivers, though it may be OK to reference a specific compression type using the new interface (e.g., saving backup history files in gz format). Unit tests only test compression using the gz format because other formats may not be available in all builds. It is the job of integration tests to exercise all compression types. Additional compression types will be added in future commits.
2020-03-06 14:41:03 -05:00
"P00 WARN: diff backup cannot alter compress-type option to 'none', reset to value in 20191003-105320F\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 INFO: execute exclusive pg_start_backup(): backup begins after the next regular checkpoint completes\n"
"P00 INFO: backup start archive = 0000000105D9759000000000, lsn = 5d97590/0\n"
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
"P00 INFO: check archive for prior segment 0000000105D9758F000000FF\n"
Add additional detail to warnings when delta checksum is auto-enabled. Hopefully this will help with debugging when it is not clear why delta checksum is being enabled.
2022-05-18 08:48:48 -04:00
"P00 WARN: file 'time-mismatch2' has timestamp (1570200100) in the future (relative to copy start 1570200000),"
" enabling delta checksum\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 WARN: resumable backup 20191003-105320F_20191004-144000D of same type exists"
" -- remove invalid files and resume\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"P00 DETAIL: remove file '" TEST_PATH "/repo/backup/test1/20191003-105320F_20191004-144000D/pg_data/PG_VERSION.gz'"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
" from resumed backup (reference in manifest)\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"P00 WARN: remove special file '" TEST_PATH "/repo/backup/test1/20191003-105320F_20191004-144000D/pg_data/pipe'"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
" from resumed backup\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"P00 DETAIL: remove file '" TEST_PATH "/repo/backup/test1/20191003-105320F_20191004-144000D/pg_data/resume-ref.gz'"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
" from resumed backup (reference in resumed manifest)\n"
Add timeline and checkpoint checks to backup. Add the following checks: * Checkpoint is updated in pg_control after pg_start_backup(). This helps ensure that PostgreSQL and pgBackRest have a consistent view of the storage and that PGDATA paths match. * Timeline of backup start WAL file matches pg_control. Hard to see how this one could get hit, but we have the power... * Standby is on the same timeline as the primary. If not, this standby is not following the primary. * Last standby checkpoint is not greater than the backup checkpoint. If so, this standby is not following the primary. This also requires some additional plumbing to read/write timeline/checkpoint from pg_control and parse timelines from WAL filenames. There were some changes in the backup tests caused by the fact that pg_control now has different contents for each backup. The check to ensure that the required checkpoint was reached on the standby should also be updated to use pg_control (it currently uses pg_control_checkpoint()), but that requires non-trivial changes to the test harness and will need to wait.
2021-12-07 09:21:07 -05:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/global/pg_control (8KB, [PCT]) checksum [SHA1]\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"P01 DETAIL: match file from prior backup " TEST_PATH "/pg1/postgresql.conf (11B, [PCT]) checksum [SHA1]\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 WARN: resumed backup file pg_data/time-mismatch2 does not have expected checksum"
Fix spacing and typos in backup/manifest modules.
2020-11-09 16:26:43 -05:00
" 984816fd329622876e14907634264e6f332e9fb3. The file will be recopied and backup will continue but this may be"
" an issue unless the resumed backup path in the repository is known to be corrupted.\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
" NOTE: this does not indicate a problem with the PostgreSQL page checksums.\n"
Change level of backup/restore copied file logging to detail. The log level for copied files in the backup/restore commands has been changed to detail. This makes the info log level less noisy but if these messages are required then set the log level for the backup/restore commands to detail.
2021-07-09 13:50:35 -04:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/time-mismatch2 (4B, [PCT]) checksum [SHA1]\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"P01 DETAIL: match file from prior backup " TEST_PATH "/pg1/PG_VERSION (3B, [PCT]) checksum [SHA1]\n"
Change level of backup/restore copied file logging to detail. The log level for copied files in the backup/restore commands has been changed to detail. This makes the info log level less noisy but if these messages are required then set the log level for the backup/restore commands to detail.
2021-07-09 13:50:35 -04:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/resume-ref (0B, [PCT])\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 DETAIL: hardlink pg_data/PG_VERSION to 20191003-105320F\n"
"P00 DETAIL: hardlink pg_data/postgresql.conf to 20191003-105320F\n"
"P00 INFO: execute exclusive pg_stop_backup() and wait for all WAL segments to archive\n"
"P00 INFO: backup stop archive = 0000000105D9759000000000, lsn = 5d97590/800000\n"
"P00 INFO: check archive for segment(s) 0000000105D9759000000000:0000000105D9759000000000\n"
Log backup file total and restore size/file total. The backup size was a bit off because it did not include any files (e.g. backup_label, WAL files) that were added to the manifest after the main copy. To fix this move the log message to the very end of the backup. Add size/file total log message to restore since it did not exist before.
2021-08-11 13:39:36 -04:00
"P00 INFO: new backup label = 20191003-105320F_20191004-144000D\n"
"P00 INFO: diff backup size = [SIZE], file total = 5");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Check repo directory
Remove last vestiges of key replacement missed in b270253a.
2021-07-15 13:19:49 -04:00
TEST_RESULT_STR_Z(
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
testBackupValidate(storageRepo(), STRDEF(STORAGE_REPO_BACKUP "/latest")),
". {link, d=20191003-105320F_20191004-144000D}\n"
"pg_data {path}\n"
"pg_data/PG_VERSION.gz {file, s=3}\n"
"pg_data/global {path}\n"
"pg_data/global/pg_control.gz {file, s=8192}\n"
"pg_data/pg_xlog {path}\n"
"pg_data/postgresql.conf.gz {file, s=11}\n"
"pg_data/resume-ref.gz {file, s=0}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"pg_data/time-mismatch2.gz {file, s=4}\n"
"--------\n"
"[backup:target]\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"pg_data={\"path\":\"" TEST_PATH "/pg1\",\"type\":\"path\"}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"\n"
"[target:file]\n"
"pg_data/PG_VERSION={\"checksum\":\"06d06bb31b570b94d7b4325f511f853dbe771c21\",\"reference\":\"20191003-105320F\""
",\"size\":3,\"timestamp\":1570000000}\n"
Add timeline and checkpoint checks to backup. Add the following checks: * Checkpoint is updated in pg_control after pg_start_backup(). This helps ensure that PostgreSQL and pgBackRest have a consistent view of the storage and that PGDATA paths match. * Timeline of backup start WAL file matches pg_control. Hard to see how this one could get hit, but we have the power... * Standby is on the same timeline as the primary. If not, this standby is not following the primary. * Last standby checkpoint is not greater than the backup checkpoint. If so, this standby is not following the primary. This also requires some additional plumbing to read/write timeline/checkpoint from pg_control and parse timelines from WAL filenames. There were some changes in the backup tests caused by the fact that pg_control now has different contents for each backup. The check to ensure that the required checkpoint was reached on the standby should also be updated to use pg_control (it currently uses pg_control_checkpoint()), but that requires non-trivial changes to the test harness and will need to wait.
2021-12-07 09:21:07 -05:00
"pg_data/global/pg_control={\"size\":8192,\"timestamp\":1570200000}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"pg_data/postgresql.conf={\"checksum\":\"e3db315c260e79211b7b52587123b7aa060f30ab\""
",\"reference\":\"20191003-105320F\",\"size\":11,\"timestamp\":1570000000}\n"
"pg_data/resume-ref={\"size\":0,\"timestamp\":1570200000}\n"
"pg_data/time-mismatch2={\"checksum\":\"984816fd329622876e14907634264e6f332e9fb3\",\"size\":4"
",\"timestamp\":1570200100}\n"
"\n"
"[target:path]\n"
"pg_data={}\n"
"pg_data/global={}\n"
"pg_data/pg_xlog={}\n",
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"compare file list");
// Remove test files
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_REMOVE(storagePgWrite(), "resume-ref", .errorOnMissing = true);
HRN_STORAGE_REMOVE(storagePgWrite(), "time-mismatch2", .errorOnMissing = true);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// -------------------------------------------------------------------------------------------------------------------------
Fix missing files corrupting the manifest. If a file was removed by PostgreSQL during the backup (or was missing from the standby) then the next file might not be copied and updated in the manifest. If this happened then the backup would error when restored. The issue was that removing files from the manifest invalidated the pointers stored in the processing queues. When a file was removed, all the pointers shifted to the next file in the list, causing a file to be unprocessed. Since the unprocessed file was still in the manifest it would be saved with no checksum, causing a failure on restore. When process-max was > 1 then the bug would often not express since the file had already been pulled from the queue and updates to the manifest are done by name rather than by pointer.
2020-01-26 13:19:13 -07:00
TEST_TITLE("online 9.6 backup-standby full backup");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
backupTimeStart = BACKUP_EPOCH + 1200000;
{
// Update pg_control
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_96);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Update version
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), PG_FILE_PGVERSION, PG_VERSION_96_STR, .timeModified = backupTimeStart);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Upgrade stanza
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRaw(argList, cfgOptRepoPath, repoPath);
hrnCfgArgRaw(argList, cfgOptPgPath, pg1Path);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptOnline, false);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdStanzaUpgrade, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
cmdStanzaUpgrade();
Replace hrnLogResult() with TEST_RESULT_LOG/_FMT(). The macros provide more information when there is an error and may be updated in the future without changing the test code.
2021-05-22 14:09:45 -04:00
TEST_RESULT_LOG("P00 INFO: stanza-upgrade for stanza 'test1' on repo1");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Load options
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRaw(argList, cfgOptRepoPath, repoPath);
hrnCfgArgKeyRaw(argList, cfgOptPgPath, 1, pg1Path);
hrnCfgArgKeyRaw(argList, cfgOptPgPath, 2, pg2Path);
hrnCfgArgKeyRawZ(argList, cfgOptPgPort, 2, "5433");
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptCompress, false);
hrnCfgArgRawBool(argList, cfgOptBackupStandby, true);
hrnCfgArgRawBool(argList, cfgOptStartFast, true);
hrnCfgArgRawBool(argList, cfgOptArchiveCopy, true);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Add timeline and checkpoint checks to backup. Add the following checks: * Checkpoint is updated in pg_control after pg_start_backup(). This helps ensure that PostgreSQL and pgBackRest have a consistent view of the storage and that PGDATA paths match. * Timeline of backup start WAL file matches pg_control. Hard to see how this one could get hit, but we have the power... * Standby is on the same timeline as the primary. If not, this standby is not following the primary. * Last standby checkpoint is not greater than the backup checkpoint. If so, this standby is not following the primary. This also requires some additional plumbing to read/write timeline/checkpoint from pg_control and parse timelines from WAL filenames. There were some changes in the backup tests caused by the fact that pg_control now has different contents for each backup. The check to ensure that the required checkpoint was reached on the standby should also be updated to use pg_control (it currently uses pg_control_checkpoint()), but that requires non-trivial changes to the test harness and will need to wait.
2021-12-07 09:21:07 -05:00
// Add pg_control to standby
HRN_PG_CONTROL_PUT(storagePgIdxWrite(1), PG_VERSION_96);
Improve testing of files that change size during the backup. Files can change size during a backup so update and add tests to cover the various scenarios more thoroughly.
2020-03-18 13:40:16 -04:00
// Create file to copy from the standby. This file will be zero-length on the primary and non-zero-length on the standby
Fix typo.
2020-03-19 12:11:20 -04:00
// but no bytes will be copied.
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_EMPTY(storagePgIdxWrite(0), PG_PATH_BASE "/1/1", .timeModified = backupTimeStart);
HRN_STORAGE_PUT_Z(storagePgIdxWrite(1), PG_PATH_BASE "/1/1", "1234");
Improve testing of files that change size during the backup. Files can change size during a backup so update and add tests to cover the various scenarios more thoroughly.
2020-03-18 13:40:16 -04:00
// Create file to copy from the standby. This file will be smaller on the primary than the standby and have no common
// data in the bytes that exist on primary and standby. If the file is copied from the primary instead of the standby
// the checksum will change but not the size.
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgIdxWrite(0), PG_PATH_BASE "/1/2", "DA", .timeModified = backupTimeStart);
HRN_STORAGE_PUT_Z(storagePgIdxWrite(1), PG_PATH_BASE "/1/2", "5678");
Improve testing of files that change size during the backup. Files can change size during a backup so update and add tests to cover the various scenarios more thoroughly.
2020-03-18 13:40:16 -04:00
// Create file to copy from the standby. This file will be larger on the primary than the standby and have no common
// data in the bytes that exist on primary and standby. If the file is copied from the primary instead of the standby
// the checksum and size will change.
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgIdxWrite(0), PG_PATH_BASE "/1/3", "TEST", .timeModified = backupTimeStart);
HRN_STORAGE_PUT_Z(storagePgIdxWrite(1), PG_PATH_BASE "/1/3", "ABC");
Fix missing files corrupting the manifest. If a file was removed by PostgreSQL during the backup (or was missing from the standby) then the next file might not be copied and updated in the manifest. If this happened then the backup would error when restored. The issue was that removing files from the manifest invalidated the pointers stored in the processing queues. When a file was removed, all the pointers shifted to the next file in the list, causing a file to be unprocessed. Since the unprocessed file was still in the manifest it would be saved with no checksum, causing a failure on restore. When process-max was > 1 then the bug would often not express since the file had already been pulled from the queue and updates to the manifest are done by name rather than by pointer.
2020-01-26 13:19:13 -07:00
// Create a file on the primary that does not exist on the standby to test that the file is removed from the manifest
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgIdxWrite(0), PG_PATH_BASE "/1/0", "DATA", .timeModified = backupTimeStart);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Set log level to warn because the following test uses multiple processes so the log order will not be deterministic
harnessLogLevelSet(logLevelWarn);
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
// Run backup but error on first archive check
testBackupPqScriptP(
PG_VERSION_96, backupTimeStart, .noPriorWal = true, .backupStandby = true, .walCompressType = compressTypeGz);
TEST_ERROR(
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
testCmdBackup(), ArchiveTimeoutError,
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
"WAL segment 0000000105DA69BF000000FF was not archived before the 100ms timeout\n"
"HINT: check the archive_command to ensure that all options are correct (especially --stanza).\n"
"HINT: check the PostgreSQL server log for errors.\n"
"HINT: run the 'start' command if the stanza was previously stopped.");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Run backup but error on archive check
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
testBackupPqScriptP(
PG_VERSION_96, backupTimeStart, .noWal = true, .backupStandby = true, .walCompressType = compressTypeGz);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_ERROR(
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
testCmdBackup(), ArchiveTimeoutError,
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"WAL segment 0000000105DA69C000000000 was not archived before the 100ms timeout\n"
"HINT: check the archive_command to ensure that all options are correct (especially --stanza).\n"
Add hint about starting the stanza when WAL segment not found. If a stop command has been issued the check command fails due to archiving timing out. Provide a hint to document this situation and point the user in the proper direction.
2020-09-03 07:49:49 -04:00
"HINT: check the PostgreSQL server log for errors.\n"
"HINT: run the 'start' command if the stanza was previously stopped.");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Remove halted backup so there's no resume
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PATH_REMOVE(storageRepoWrite(), STORAGE_REPO_BACKUP "/20191016-042640F", .recurse = true);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Run backup
Add infrastructure for multiple compression type support. Add compress-type option and deprecate compress option. Since the compress option is boolean it won't work with multiple compression types. Add logic to cfgLoadUpdateOption() to update compress-type if it is not set directly. The compress option should no longer be referenced outside the cfgLoadUpdateOption() function. Add common/compress/helper module to contain interface functions that work with multiple compression types. Code outside this module should no longer call specific compression drivers, though it may be OK to reference a specific compression type using the new interface (e.g., saving backup history files in gz format). Unit tests only test compression using the gz format because other formats may not be available in all builds. It is the job of integration tests to exercise all compression types. Additional compression types will be added in future commits.
2020-03-06 14:41:03 -05:00
testBackupPqScriptP(PG_VERSION_96, backupTimeStart, .backupStandby = true, .walCompressType = compressTypeGz);
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Set log level back to detail
harnessLogLevelSet(logLevelDetail);
TEST_RESULT_LOG(
"P00 WARN: no prior backup exists, incr backup has been changed to full");
Remove last vestiges of key replacement missed in b270253a.
2021-07-15 13:19:49 -04:00
TEST_RESULT_STR_Z(
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
testBackupValidate(storageRepo(), STRDEF(STORAGE_REPO_BACKUP "/latest")),
". {link, d=20191016-042640F}\n"
"pg_data {path}\n"
"pg_data/PG_VERSION {file, s=3}\n"
"pg_data/backup_label {file, s=17}\n"
"pg_data/base {path}\n"
"pg_data/base/1 {path}\n"
Limit backup file copy size to size reported at backup start. If a file grows during the backup it will be reconstructed by WAL replay during recovery so there is no need to copy the additional data. This also reduces the likelihood of seeing torn pages during the copy. Torn pages can still occur in the middle of the file, though, so they must be handled.
2020-03-19 13:16:05 -04:00
"pg_data/base/1/1 {file, s=0}\n"
"pg_data/base/1/2 {file, s=2}\n"
Improve testing of files that change size during the backup. Files can change size during a backup so update and add tests to cover the various scenarios more thoroughly.
2020-03-18 13:40:16 -04:00
"pg_data/base/1/3 {file, s=3}\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"pg_data/global {path}\n"
"pg_data/global/pg_control {file, s=8192}\n"
"pg_data/pg_xlog {path}\n"
"pg_data/pg_xlog/0000000105DA69C000000000 {file, s=16777216}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"pg_data/postgresql.conf {file, s=11}\n"
"--------\n"
"[backup:target]\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"pg_data={\"path\":\"" TEST_PATH "/pg1\",\"type\":\"path\"}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"\n"
"[target:file]\n"
"pg_data/PG_VERSION={\"checksum\":\"f5b7e6d36dc0113f61b36c700817d42b96f7b037\",\"size\":3"
",\"timestamp\":1571200000}\n"
"pg_data/backup_label={\"checksum\":\"8e6f41ac87a7514be96260d65bacbffb11be77dc\",\"size\":17"
",\"timestamp\":1571200002}\n"
Remove primary flag from manifest. This flag was only being used by the backup command after manifestNewBuild() and had no other uses. There was a time when it was important for integration testing but the unit tests now fulfill this role. Since backup is the only code concerned with the primary flag, move the code into the backup module. We don't have any cross-version testing but this change was tested manually with the most recent version of pgBackRest to make sure it was tolerant of the missing primary info. When an older version of pgBackRest loads a newer manifest the primary flag will always be set to false, which is fine since it is not used.
2022-01-20 14:01:10 -05:00
"pg_data/base/1/1={\"size\":0,\"timestamp\":1571200000}\n"
"pg_data/base/1/2={\"checksum\":\"54ceb91256e8190e474aa752a6e0650a2df5ba37\",\"size\":2,\"timestamp\":1571200000}\n"
"pg_data/base/1/3={\"checksum\":\"3c01bdbb26f358bab27f267924aa2c9a03fcfdb8\",\"size\":3,\"timestamp\":1571200000}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"pg_data/global/pg_control={\"size\":8192,\"timestamp\":1571200000}\n"
"pg_data/pg_xlog/0000000105DA69C000000000={\"size\":16777216,\"timestamp\":1571200002}\n"
"pg_data/postgresql.conf={\"checksum\":\"e3db315c260e79211b7b52587123b7aa060f30ab\",\"size\":11"
",\"timestamp\":1570000000}\n"
"\n"
"[target:path]\n"
"pg_data={}\n"
"pg_data/base={}\n"
"pg_data/base/1={}\n"
"pg_data/global={}\n"
"pg_data/pg_xlog={}\n",
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"compare file list");
// Remove test files
Allow NULL path in HRN_STORAGE_PATH_REMOVE() macro. Update command/backup test to pass NULL where appropriate.
2021-07-19 15:20:43 -04:00
HRN_STORAGE_PATH_REMOVE(storagePgIdxWrite(1), NULL, .recurse = true);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PATH_REMOVE(storagePgWrite(), "base/1", .recurse = true);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("online 11 full backup with tablespaces and page checksums");
backupTimeStart = BACKUP_EPOCH + 2200000;
{
// Update pg_control
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
HRN_PG_CONTROL_PUT(storagePgWrite(), PG_VERSION_11, .pageChecksum = true, .walSegmentSize = 1024 * 1024);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Update version
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_Z(storagePgWrite(), PG_FILE_PGVERSION, PG_VERSION_11_STR, .timeModified = backupTimeStart);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Update wal path
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PATH_REMOVE(storagePgWrite(), strZ(pgWalPath(PG_VERSION_95)));
HRN_STORAGE_PATH_CREATE(storagePgWrite(), strZ(pgWalPath(PG_VERSION_11)), .noParentCreate = true);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Upgrade stanza
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRaw(argList, cfgOptRepoPath, repoPath);
hrnCfgArgRaw(argList, cfgOptPgPath, pg1Path);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptOnline, false);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdStanzaUpgrade, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
cmdStanzaUpgrade();
Replace hrnLogResult() with TEST_RESULT_LOG/_FMT(). The macros provide more information when there is an error and may be updated in the future without changing the test code.
2021-05-22 14:09:45 -04:00
TEST_RESULT_LOG("P00 INFO: stanza-upgrade for stanza 'test1' on repo1");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Load options
argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRaw(argList, cfgOptRepoPath, repoPath);
hrnCfgArgRaw(argList, cfgOptPgPath, pg1Path);
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Convert BackupType enum to StringId. Allows removal of backupType()/backupTypeStr() and improves debug logging of the enum. Move BackupType enum and string constants to info/infoBackup.h so they are available to more modules. Also convert InfoBackup to use BackupType instead of a String.
2021-05-03 12:15:39 -04:00
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeFull);
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawBool(argList, cfgOptRepoHardlink, true);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptManifestSaveThreshold, "1");
hrnCfgArgRawBool(argList, cfgOptArchiveCopy, true);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Restore works when PGDATA is a link. Make the restore clean process look more like manifest build, i.e. do cleanup of each target root directory outside the main cleanup callback. This means some code duplication but removes the logic handling "dot" paths. Add tests for both restore and backup (which already worked but was not tested).
2020-04-21 17:55:36 -04:00
// Move pg1-path and put a link in its place. This tests that backup works when pg1-path is a symlink yet should be
// completely invisible in the manifest and logging.
Rename TEST_SYSTEM*() to HRN_SYSTEM*(). These calls are not tests, rather they setup data for tests.
2021-05-22 14:22:51 -04:00
HRN_SYSTEM_FMT("mv %s %s-data", strZ(pg1Path), strZ(pg1Path));
HRN_SYSTEM_FMT("ln -s %s-data %s ", strZ(pg1Path), strZ(pg1Path));
Restore works when PGDATA is a link. Make the restore clean process look more like manifest build, i.e. do cleanup of each target root directory outside the main cleanup callback. This means some code duplication but removes the logic handling "dot" paths. Add tests for both restore and backup (which already worked but was not tested).
2020-04-21 17:55:36 -04:00
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Zeroed file which passes page checksums
Buffer *relation = bufNew(PG_PAGE_SIZE_DEFAULT);
memset(bufPtr(relation), 0, bufSize(relation));
bufUsedSet(relation, bufSize(relation));
Use designated initializers to initialize structs. Previously memNew() used memset() to initialize all struct members to 0, NULL, false, etc. While this appears to work in practice, it is a violation of the C specification. For instance, NULL == 0 must be true but neither NULL nor 0 must be represented with all zero bits. Instead use designated initializers to initialize structs. These guarantee that struct members will be properly initialized even if they are not specified in the initializer. Note that due to a quirk in the C99 specification at least one member must be explicitly initialized even if it needs to be the default value. Since pre-zeroed memory is no longer required, adjust memAllocInternal()/memReallocInternal() to return raw memory and update dependent functions accordingly. All instances of memset() have been removed except in debug/test code where needed. Add memMewPtrArray() to allocate an array of pointers and automatically set all pointers to NULL. Rename memGrowRaw() to the more logical memResize().
2020-01-23 14:15:58 -07:00
*(PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x00)) = (PageHeaderData){.pd_upper = 0};
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT(storagePgWrite(), PG_PATH_BASE "/1/1", relation, .timeModified = backupTimeStart);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
// File which will fail on alignment
relation = bufNew(PG_PAGE_SIZE_DEFAULT + 512);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
memset(bufPtr(relation), 0, bufSize(relation));
bufUsedSet(relation, bufSize(relation));
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
*(PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x00)) = (PageHeaderData){.pd_upper = 0xFE};
((PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x00)))->pd_checksum = pgPageChecksum(
bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x00), 0);
*(PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x01)) = (PageHeaderData){.pd_upper = 0xFF};
Use designated initializers to initialize structs. Previously memNew() used memset() to initialize all struct members to 0, NULL, false, etc. While this appears to work in practice, it is a violation of the C specification. For instance, NULL == 0 must be true but neither NULL nor 0 must be represented with all zero bits. Instead use designated initializers to initialize structs. These guarantee that struct members will be properly initialized even if they are not specified in the initializer. Note that due to a quirk in the C99 specification at least one member must be explicitly initialized even if it needs to be the default value. Since pre-zeroed memory is no longer required, adjust memAllocInternal()/memReallocInternal() to return raw memory and update dependent functions accordingly. All instances of memset() have been removed except in debug/test code where needed. Add memMewPtrArray() to allocate an array of pointers and automatically set all pointers to NULL. Rename memGrowRaw() to the more logical memResize().
2020-01-23 14:15:58 -07:00
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT(storagePgWrite(), PG_PATH_BASE "/1/2", relation, .timeModified = backupTimeStart);
Use StringId for type in cryptoHash() and cryptoHmacOne(). This brings hash types in line with cipher types (i.e. a StringId enum) and allows removal of some Strings.
2022-05-25 15:27:53 -04:00
const char *rel1_2Sha1 = strZ(bufHex(cryptoHashOne(hashTypeSha1, relation)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// File with bad page checksums
Improve validation of zero pages. Checking that pd_upper == 0 is not enough since this field may be corrupted. Still use pd_upper as a quick check, but when it is zero proceed to check the rest of the page to ensure it is also all zeroes.
2022-02-23 13:17:14 -06:00
relation = bufNew(PG_PAGE_SIZE_DEFAULT * 5);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
memset(bufPtr(relation), 0, bufSize(relation));
Use designated initializers to initialize structs. Previously memNew() used memset() to initialize all struct members to 0, NULL, false, etc. While this appears to work in practice, it is a violation of the C specification. For instance, NULL == 0 must be true but neither NULL nor 0 must be represented with all zero bits. Instead use designated initializers to initialize structs. These guarantee that struct members will be properly initialized even if they are not specified in the initializer. Note that due to a quirk in the C99 specification at least one member must be explicitly initialized even if it needs to be the default value. Since pre-zeroed memory is no longer required, adjust memAllocInternal()/memReallocInternal() to return raw memory and update dependent functions accordingly. All instances of memset() have been removed except in debug/test code where needed. Add memMewPtrArray() to allocate an array of pointers and automatically set all pointers to NULL. Rename memGrowRaw() to the more logical memResize().
2020-01-23 14:15:58 -07:00
*(PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x00)) = (PageHeaderData){.pd_upper = 0xFF};
*(PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x01)) = (PageHeaderData){.pd_upper = 0x00};
*(PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x02)) = (PageHeaderData){.pd_upper = 0xFE};
*(PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x03)) = (PageHeaderData){.pd_upper = 0xEF};
Improve validation of zero pages. Checking that pd_upper == 0 is not enough since this field may be corrupted. Still use pd_upper as a quick check, but when it is zero proceed to check the rest of the page to ensure it is also all zeroes.
2022-02-23 13:17:14 -06:00
*(PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x04)) = (PageHeaderData){.pd_upper = 0x00};
(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x04))[PG_PAGE_SIZE_DEFAULT - 1] = 0xFF;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
bufUsedSet(relation, bufSize(relation));
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT(storagePgWrite(), PG_PATH_BASE "/1/3", relation, .timeModified = backupTimeStart);
Use StringId for type in cryptoHash() and cryptoHmacOne(). This brings hash types in line with cipher types (i.e. a StringId enum) and allows removal of some Strings.
2022-05-25 15:27:53 -04:00
const char *rel1_3Sha1 = strZ(bufHex(cryptoHashOne(hashTypeSha1, relation)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// File with bad page checksum
relation = bufNew(PG_PAGE_SIZE_DEFAULT * 3);
memset(bufPtr(relation), 0, bufSize(relation));
Use designated initializers to initialize structs. Previously memNew() used memset() to initialize all struct members to 0, NULL, false, etc. While this appears to work in practice, it is a violation of the C specification. For instance, NULL == 0 must be true but neither NULL nor 0 must be represented with all zero bits. Instead use designated initializers to initialize structs. These guarantee that struct members will be properly initialized even if they are not specified in the initializer. Note that due to a quirk in the C99 specification at least one member must be explicitly initialized even if it needs to be the default value. Since pre-zeroed memory is no longer required, adjust memAllocInternal()/memReallocInternal() to return raw memory and update dependent functions accordingly. All instances of memset() have been removed except in debug/test code where needed. Add memMewPtrArray() to allocate an array of pointers and automatically set all pointers to NULL. Rename memGrowRaw() to the more logical memResize().
2020-01-23 14:15:58 -07:00
*(PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x00)) = (PageHeaderData){.pd_upper = 0x00};
*(PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x01)) = (PageHeaderData){.pd_upper = 0x08};
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
*(PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x02)) = (PageHeaderData){.pd_upper = 0xFF};
((PageHeaderData *)(bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x02)))->pd_checksum = pgPageChecksum(
bufPtr(relation) + (PG_PAGE_SIZE_DEFAULT * 0x02), 2);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
bufUsedSet(relation, bufSize(relation));
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT(storagePgWrite(), PG_PATH_BASE "/1/4", relation, .timeModified = backupTimeStart);
Use StringId for type in cryptoHash() and cryptoHmacOne(). This brings hash types in line with cipher types (i.e. a StringId enum) and allows removal of some Strings.
2022-05-25 15:27:53 -04:00
const char *rel1_4Sha1 = strZ(bufHex(cryptoHashOne(hashTypeSha1, relation)));
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Add a tablespace
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PATH_CREATE(storagePgWrite(), PG_PATH_PGTBLSPC);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
THROW_ON_SYS_ERROR(
Rename strPtr()/strPtrNull() to strZ()/strZNull(). We use the Z suffix in many functions to indicate that we are expecting a zero-terminated string so make this function conform to the pattern. As a bonus the new name is a bit shorter, which is a good quality in a commonly-used function.
2020-07-30 07:49:06 -04:00
symlink("../../pg1-tblspc/32768", strZ(storagePathP(storagePg(), STRDEF(PG_PATH_PGTBLSPC "/32768")))) == -1,
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
FileOpenError, "unable to create symlink");
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PUT_EMPTY(
storageTest,
Add zNewFmt(). This replaces strZ(strNewFmt()), making the code simpler and reducing indentation.
2022-05-06 12:32:49 -04:00
zNewFmt("pg1-tblspc/32768/%s/1/5", strZ(pgTablespaceId(PG_VERSION_11, hrnPgCatalogVersion(PG_VERSION_11)))),
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
.timeModified = backupTimeStart);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Disable storageFeatureSymLink so tablespace (and latest) symlinks will not be created
Refactor storage modules with inline getters/setters. Extended the pattern introduced in 79a2d02c to the storage modules: Storage, StorageRead, StorageWrite.
2021-04-07 14:04:38 -04:00
((Storage *)storageRepoWrite())->pub.interface.feature ^= 1 << storageFeatureSymLink;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Disable storageFeatureHardLink so hardlinks will not be created
Refactor storage modules with inline getters/setters. Extended the pattern introduced in 79a2d02c to the storage modules: Storage, StorageRead, StorageWrite.
2021-04-07 14:04:38 -04:00
((Storage *)storageRepoWrite())->pub.interface.feature ^= 1 << storageFeatureHardLink;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Run backup
Add infrastructure for multiple compression type support. Add compress-type option and deprecate compress option. Since the compress option is boolean it won't work with multiple compression types. Add logic to cfgLoadUpdateOption() to update compress-type if it is not set directly. The compress option should no longer be referenced outside the cfgLoadUpdateOption() function. Add common/compress/helper module to contain interface functions that work with multiple compression types. Code outside this module should no longer call specific compression drivers, though it may be OK to reference a specific compression type using the new interface (e.g., saving backup history files in gz format). Unit tests only test compression using the gz format because other formats may not be available in all builds. It is the job of integration tests to exercise all compression types. Additional compression types will be added in future commits.
2020-03-06 14:41:03 -05:00
testBackupPqScriptP(PG_VERSION_11, backupTimeStart, .walCompressType = compressTypeGz, .walTotal = 3);
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Reset storage features
Refactor storage modules with inline getters/setters. Extended the pattern introduced in 79a2d02c to the storage modules: Storage, StorageRead, StorageWrite.
2021-04-07 14:04:38 -04:00
((Storage *)storageRepoWrite())->pub.interface.feature |= 1 << storageFeatureSymLink;
((Storage *)storageRepoWrite())->pub.interface.feature |= 1 << storageFeatureHardLink;
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_LOG(
"P00 INFO: execute non-exclusive pg_start_backup(): backup begins after the next regular checkpoint completes\n"
"P00 INFO: backup start archive = 0000000105DB5DE000000000, lsn = 5db5de0/0\n"
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
"P00 INFO: check archive for segment 0000000105DB5DE000000000\n"
Improve validation of zero pages. Checking that pd_upper == 0 is not enough since this field may be corrupted. Still use pd_upper as a quick check, but when it is zero proceed to check the rest of the page to ensure it is also all zeroes.
2022-02-23 13:17:14 -06:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/base/1/3 (40KB, [PCT]) checksum [SHA1]\n"
"P00 WARN: invalid page checksums found in file " TEST_PATH "/pg1/base/1/3 at pages 0, 2-4\n"
Change level of backup/restore copied file logging to detail. The log level for copied files in the backup/restore commands has been changed to detail. This makes the info log level less noisy but if these messages are required then set the log level for the backup/restore commands to detail.
2021-07-09 13:50:35 -04:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/base/1/4 (24KB, [PCT]) checksum [SHA1]\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"P00 WARN: invalid page checksum found in file " TEST_PATH "/pg1/base/1/4 at page 1\n"
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/base/1/2 (8.5KB, [PCT]) checksum [SHA1]\n"
"P00 WARN: page misalignment in file " TEST_PATH "/pg1/base/1/2: file size 8704 is not divisible by page size"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
" 8192\n"
Change level of backup/restore copied file logging to detail. The log level for copied files in the backup/restore commands has been changed to detail. This makes the info log level less noisy but if these messages are required then set the log level for the backup/restore commands to detail.
2021-07-09 13:50:35 -04:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/global/pg_control (8KB, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/base/1/1 (8KB, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/postgresql.conf (11B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/PG_VERSION (2B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/pg_tblspc/32768/PG_11_201809051/1/5 (0B, [PCT])\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 INFO: execute non-exclusive pg_stop_backup() and wait for all WAL segments to archive\n"
"P00 INFO: backup stop archive = 0000000105DB5DE000000002, lsn = 5db5de0/280000\n"
"P00 DETAIL: wrote 'backup_label' file returned from pg_stop_backup()\n"
Add tablespace_map file to command/backup test module. The code worked fine but better to have explicit tests for this file.
2021-11-15 14:32:22 -05:00
"P00 DETAIL: wrote 'tablespace_map' file returned from pg_stop_backup()\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 INFO: check archive for segment(s) 0000000105DB5DE000000000:0000000105DB5DE000000002\n"
Log archive copy during backup. Copying can be a fairly expensive operation so it makes sense to log it so the user gets some status during long copy operations.
2021-03-11 08:22:44 -05:00
"P00 DETAIL: copy segment 0000000105DB5DE000000000 to backup\n"
"P00 DETAIL: copy segment 0000000105DB5DE000000001 to backup\n"
"P00 DETAIL: copy segment 0000000105DB5DE000000002 to backup\n"
Log backup file total and restore size/file total. The backup size was a bit off because it did not include any files (e.g. backup_label, WAL files) that were added to the manifest after the main copy. To fix this move the log message to the very end of the backup. Add size/file total log message to restore since it did not exist before.
2021-08-11 13:39:36 -04:00
"P00 INFO: new backup label = 20191027-181320F\n"
Add tablespace_map file to command/backup test module. The code worked fine but better to have explicit tests for this file.
2021-11-15 14:32:22 -05:00
"P00 INFO: full backup size = [SIZE], file total = 13");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Remove last vestiges of key replacement missed in b270253a.
2021-07-15 13:19:49 -04:00
TEST_RESULT_STR(
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
testBackupValidate(storageRepo(), STRDEF(STORAGE_REPO_BACKUP "/20191027-181320F")),
Reduce reliance on static checksums in unit tests. Testing against static checksums is valuable but it can be become burdensome when supporting multiple architectures. Reduce the number of tests we are doing against static checksums when the architecture can cause the checksum to vary.
2020-07-20 09:47:43 -04:00
strNewFmt(
"pg_data {path}\n"
"pg_data/PG_VERSION.gz {file, s=2}\n"
"pg_data/backup_label.gz {file, s=17}\n"
"pg_data/base {path}\n"
"pg_data/base/1 {path}\n"
"pg_data/base/1/1.gz {file, s=8192}\n"
Retry on page checksum validation failure during backup. Rather than attempting to filter page checksum failures by LSN, just retry when there is a page checksum failure. If the page has not changed since the last read report it as an error. If the page has changed, then PostgreSQL must be modifying the page so we can ignore the error because a full page write (and possibly updates) will be in the WAL. Also remove tests made redundant by the test merge in b4897077.
2022-02-23 12:05:53 -06:00
"pg_data/base/1/2.gz {file, s=8704}\n"
Improve validation of zero pages. Checking that pd_upper == 0 is not enough since this field may be corrupted. Still use pd_upper as a quick check, but when it is zero proceed to check the rest of the page to ensure it is also all zeroes.
2022-02-23 13:17:14 -06:00
"pg_data/base/1/3.gz {file, s=40960}\n"
Reduce reliance on static checksums in unit tests. Testing against static checksums is valuable but it can be become burdensome when supporting multiple architectures. Reduce the number of tests we are doing against static checksums when the architecture can cause the checksum to vary.
2020-07-20 09:47:43 -04:00
"pg_data/base/1/4.gz {file, s=24576}\n"
"pg_data/global {path}\n"
"pg_data/global/pg_control.gz {file, s=8192}\n"
"pg_data/pg_tblspc {path}\n"
"pg_data/pg_wal {path}\n"
"pg_data/pg_wal/0000000105DB5DE000000000.gz {file, s=1048576}\n"
"pg_data/pg_wal/0000000105DB5DE000000001.gz {file, s=1048576}\n"
"pg_data/pg_wal/0000000105DB5DE000000002.gz {file, s=1048576}\n"
"pg_data/postgresql.conf.gz {file, s=11}\n"
Add tablespace_map file to command/backup test module. The code worked fine but better to have explicit tests for this file.
2021-11-15 14:32:22 -05:00
"pg_data/tablespace_map.gz {file, s=19}\n"
Reduce reliance on static checksums in unit tests. Testing against static checksums is valuable but it can be become burdensome when supporting multiple architectures. Reduce the number of tests we are doing against static checksums when the architecture can cause the checksum to vary.
2020-07-20 09:47:43 -04:00
"pg_tblspc {path}\n"
"pg_tblspc/32768 {path}\n"
"pg_tblspc/32768/PG_11_201809051 {path}\n"
"pg_tblspc/32768/PG_11_201809051/1 {path}\n"
"pg_tblspc/32768/PG_11_201809051/1/5.gz {file, s=0}\n"
"--------\n"
"[backup:target]\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"pg_data={\"path\":\"" TEST_PATH "/pg1\",\"type\":\"path\"}\n"
Reduce reliance on static checksums in unit tests. Testing against static checksums is valuable but it can be become burdensome when supporting multiple architectures. Reduce the number of tests we are doing against static checksums when the architecture can cause the checksum to vary.
2020-07-20 09:47:43 -04:00
"pg_tblspc/32768={\"path\":\"../../pg1-tblspc/32768\",\"tablespace-id\":\"32768\""
",\"tablespace-name\":\"tblspc32768\",\"type\":\"link\"}\n"
"\n"
"[target:file]\n"
"pg_data/PG_VERSION={\"checksum\":\"17ba0791499db908433b80f37c5fbc89b870084b\",\"size\":2"
",\"timestamp\":1572200000}\n"
"pg_data/backup_label={\"checksum\":\"8e6f41ac87a7514be96260d65bacbffb11be77dc\",\"size\":17"
",\"timestamp\":1572200002}\n"
"pg_data/base/1/1={\"checksum\":\"0631457264ff7f8d5fb1edc2c0211992a67c73e6\",\"checksum-page\":true"
Remove primary flag from manifest. This flag was only being used by the backup command after manifestNewBuild() and had no other uses. There was a time when it was important for integration testing but the unit tests now fulfill this role. Since backup is the only code concerned with the primary flag, move the code into the backup module. We don't have any cross-version testing but this change was tested manually with the most recent version of pgBackRest to make sure it was tolerant of the missing primary info. When an older version of pgBackRest loads a newer manifest the primary flag will always be set to false, which is fine since it is not used.
2022-01-20 14:01:10 -05:00
",\"size\":8192,\"timestamp\":1572200000}\n"
Add limited CI for ppc64le/s390x using emulation. The emulation is so slow that running all the unit tests would be too expensive, but this at least shows that the build works and some of the more complex tests run. In particular, it is good to test on one big-endian architecture to be sure that checksums are correct. Update checksums in the tests where they had gotten out of date since the last time we were testing on s390x. Also use a different test in command/archivePushTest to show the name of the file when a checksum does not match to aid in debugging. The command/archive-push test was updated but not included because there is also a permissions issue, which looks to be the same as what we see on MacOS/FreedBSD. Hopefully we'll be able to fix all of those at the same time.
2022-05-09 12:48:19 -04:00
"pg_data/base/1/2={\"checksum\":\"%s\",\"checksum-page\":false,\"size\":8704,\"timestamp\":1572200000}\n"
Improve validation of zero pages. Checking that pd_upper == 0 is not enough since this field may be corrupted. Still use pd_upper as a quick check, but when it is zero proceed to check the rest of the page to ensure it is also all zeroes.
2022-02-23 13:17:14 -06:00
"pg_data/base/1/3={\"checksum\":\"%s\",\"checksum-page\":false,\"checksum-page-error\":[0,[2,4]]"
",\"size\":40960,\"timestamp\":1572200000}\n"
Remove primary flag from manifest. This flag was only being used by the backup command after manifestNewBuild() and had no other uses. There was a time when it was important for integration testing but the unit tests now fulfill this role. Since backup is the only code concerned with the primary flag, move the code into the backup module. We don't have any cross-version testing but this change was tested manually with the most recent version of pgBackRest to make sure it was tolerant of the missing primary info. When an older version of pgBackRest loads a newer manifest the primary flag will always be set to false, which is fine since it is not used.
2022-01-20 14:01:10 -05:00
"pg_data/base/1/4={\"checksum\":\"%s\",\"checksum-page\":false,\"checksum-page-error\":[1],\"size\":24576"
",\"timestamp\":1572200000}\n"
Reduce reliance on static checksums in unit tests. Testing against static checksums is valuable but it can be become burdensome when supporting multiple architectures. Reduce the number of tests we are doing against static checksums when the architecture can cause the checksum to vary.
2020-07-20 09:47:43 -04:00
"pg_data/global/pg_control={\"size\":8192,\"timestamp\":1572200000}\n"
"pg_data/pg_wal/0000000105DB5DE000000000={\"size\":1048576,\"timestamp\":1572200002}\n"
"pg_data/pg_wal/0000000105DB5DE000000001={\"size\":1048576,\"timestamp\":1572200002}\n"
"pg_data/pg_wal/0000000105DB5DE000000002={\"size\":1048576,\"timestamp\":1572200002}\n"
"pg_data/postgresql.conf={\"checksum\":\"e3db315c260e79211b7b52587123b7aa060f30ab\",\"size\":11"
",\"timestamp\":1570000000}\n"
Add tablespace_map file to command/backup test module. The code worked fine but better to have explicit tests for this file.
2021-11-15 14:32:22 -05:00
"pg_data/tablespace_map={\"checksum\":\"87fe624d7976c2144e10afcb7a9a49b071f35e9c\",\"size\":19"
",\"timestamp\":1572200002}\n"
Remove primary flag from manifest. This flag was only being used by the backup command after manifestNewBuild() and had no other uses. There was a time when it was important for integration testing but the unit tests now fulfill this role. Since backup is the only code concerned with the primary flag, move the code into the backup module. We don't have any cross-version testing but this change was tested manually with the most recent version of pgBackRest to make sure it was tolerant of the missing primary info. When an older version of pgBackRest loads a newer manifest the primary flag will always be set to false, which is fine since it is not used.
2022-01-20 14:01:10 -05:00
"pg_tblspc/32768/PG_11_201809051/1/5={\"checksum-page\":true,\"size\":0,\"timestamp\":1572200000}\n"
Reduce reliance on static checksums in unit tests. Testing against static checksums is valuable but it can be become burdensome when supporting multiple architectures. Reduce the number of tests we are doing against static checksums when the architecture can cause the checksum to vary.
2020-07-20 09:47:43 -04:00
"\n"
"[target:link]\n"
"pg_data/pg_tblspc/32768={\"destination\":\"../../pg1-tblspc/32768\"}\n"
"\n"
"[target:path]\n"
"pg_data={}\n"
"pg_data/base={}\n"
"pg_data/base/1={}\n"
"pg_data/global={}\n"
"pg_data/pg_tblspc={}\n"
"pg_data/pg_wal={}\n"
"pg_tblspc={}\n"
"pg_tblspc/32768={}\n"
"pg_tblspc/32768/PG_11_201809051={}\n"
"pg_tblspc/32768/PG_11_201809051/1={}\n",
Add limited CI for ppc64le/s390x using emulation. The emulation is so slow that running all the unit tests would be too expensive, but this at least shows that the build works and some of the more complex tests run. In particular, it is good to test on one big-endian architecture to be sure that checksums are correct. Update checksums in the tests where they had gotten out of date since the last time we were testing on s390x. Also use a different test in command/archivePushTest to show the name of the file when a checksum does not match to aid in debugging. The command/archive-push test was updated but not included because there is also a permissions issue, which looks to be the same as what we see on MacOS/FreedBSD. Hopefully we'll be able to fix all of those at the same time.
2022-05-09 12:48:19 -04:00
rel1_2Sha1, rel1_3Sha1, rel1_4Sha1),
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"compare file list");
// Remove test files
Add optional remove to TEST_STORAGE_EXISTS(). This allows TEST_STORAGE_EXISTS() to be used in most cases where TEST_STORAGE_REMOVE() was used before. Rename TEST_STORAGE_REMOVE() to HRN_STORAGE_REMOVE() now that is is no longer used as a test. Still allow an error when the file is missing just to help keep tests tidy.
2021-06-08 14:51:23 -04:00
HRN_STORAGE_REMOVE(storagePgWrite(), "base/1/2", .errorOnMissing = true);
HRN_STORAGE_REMOVE(storagePgWrite(), "base/1/3", .errorOnMissing = true);
HRN_STORAGE_REMOVE(storagePgWrite(), "base/1/4", .errorOnMissing = true);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("error when pg_control not present");
{
// Load options
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRaw(argList, cfgOptRepoPath, repoPath);
hrnCfgArgRaw(argList, cfgOptPgPath, pg1Path);
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Convert BackupType enum to StringId. Allows removal of backupType()/backupTypeStr() and improves debug logging of the enum. Move BackupType enum and string constants to info/infoBackup.h so they are available to more modules. Also convert InfoBackup to use BackupType instead of a String.
2021-05-03 12:15:39 -04:00
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeIncr);
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawBool(argList, cfgOptRepoHardlink, true);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Add HRN_PG_CONTROL_PUT() and HRN_PG_CONTROL_TIME(). These macros simplify management of pg_control test files. Centralize time updates for pg_control in the command/backup module. This caused some time updates in the logs. Finally, move the postgres module after the storage module so it can use storage macros.
2021-11-30 13:23:11 -05:00
// Preserve prior timestamp on pg_control
testBackupPqScriptP(PG_VERSION_11, BACKUP_EPOCH + 2300000, .errorAfterStart = true);
HRN_PG_CONTROL_TIME(storagePg(), backupTimeStart);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Run backup
TEST_ERROR(
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
testCmdBackup(), FileMissingError,
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"pg_control must be present in all online backups\n"
"HINT: is something wrong with the clock or filesystem timestamps?");
// Check log
TEST_RESULT_LOG(
"P00 INFO: last backup label = 20191027-181320F, version = " PROJECT_VERSION "\n"
"P00 INFO: execute non-exclusive pg_start_backup(): backup begins after the next regular checkpoint completes\n"
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
"P00 INFO: backup start archive = 0000000105DB764000000000, lsn = 5db7640/0\n"
"P00 INFO: check archive for prior segment 0000000105DB763F00000FFF");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
// Remove partial backup so it won't be resumed (since it errored before any checksums were written)
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
HRN_STORAGE_PATH_REMOVE(storageRepoWrite(), STORAGE_REPO_BACKUP "/20191027-181320F_20191028-220000I", .recurse = true);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("online 11 incr backup with tablespaces");
backupTimeStart = BACKUP_EPOCH + 2400000;
{
// Load options
StringList *argList = strLstNew();
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
Partial multi-repository implementation. Multi-repository implementations for the archive-push, check, info, stanza-create, stanza-upgrade, and stanza-delete commands. Multi-repo configuration is disabled so there should be no behavioral changes between these commands and their current single-repo implementations. Multi-repo documentation and integration tests are still in the multi-repo development branch. All unit tests work as multi-repo since they are able to bypass the configuration restrictions.
2021-01-21 15:21:50 -05:00
hrnCfgArgKeyRawZ(argList, cfgOptRepoPath, 1, "/repo-bogus");
hrnCfgArgKeyRaw(argList, cfgOptRepoPath, 2, repoPath);
hrnCfgArgKeyRawZ(argList, cfgOptRepoRetentionFull, 2, "1");
hrnCfgArgKeyRawBool(argList, cfgOptRepoHardlink, 2, true);
hrnCfgArgRawZ(argList, cfgOptRepo, "2");
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRaw(argList, cfgOptPgPath, pg1Path);
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
Convert BackupType enum to StringId. Allows removal of backupType()/backupTypeStr() and improves debug logging of the enum. Move BackupType enum and string constants to info/infoBackup.h so they are available to more modules. Also convert InfoBackup to use BackupType instead of a String.
2021-05-03 12:15:39 -04:00
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeIncr);
Update command/backup tests to use standard patterns. Includes backup and backupCommon tests. Some tests in backupTest were split out where they were originally combined into a single boolean check - which made it difficult to determine which part of the conditional failed. String values were also removed where they were no longer needed.
2021-07-15 17:00:20 -04:00
hrnCfgArgRawBool(argList, cfgOptDelta, true);
Remove indexed option constants. These constants don't scale well as the index total is increased for an option. The core code rarely uses these options and they are easily replaced with cfgOptionName(). The tests had started to make use of the constants, so provide functions that build the option name from the optionId and, optionally, the optionKey.
2020-10-19 14:03:48 -04:00
hrnCfgArgRawBool(argList, cfgOptRepoHardlink, true);
Replace harnessCfgLoad*() functions with HRN_CFG_LOAD() macro. HRN_CFG_LOAD() handles the majority of test configuration loads and has various options for special cases. It was not clear when to use harnessCfgLoadRaw() vs harnessCfgLoad(). Now "raw" functionality is granular and enabled by parameters, e.g. noStd.
2021-06-01 09:03:44 -04:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Fix error in timeline conversion. The timeline is required to verify WAL segments in the archive after a backup. The conversion was performed base 10 instead of 16, which led to errors when the timeline was ≥ 0xA.
2020-01-21 10:29:46 -07:00
// Run backup. Make sure that the timeline selected converts to hexdecimal that can't be interpreted as decimal.
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
testBackupPqScriptP(PG_VERSION_11, backupTimeStart, .timeline = 0x2C, .walTotal = 2);
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
TEST_RESULT_LOG(
"P00 INFO: last backup label = 20191027-181320F, version = " PROJECT_VERSION "\n"
"P00 INFO: execute non-exclusive pg_start_backup(): backup begins after the next regular checkpoint completes\n"
Fix error in timeline conversion. The timeline is required to verify WAL segments in the archive after a backup. The conversion was performed base 10 instead of 16, which led to errors when the timeline was ≥ 0xA.
2020-01-21 10:29:46 -07:00
"P00 INFO: backup start archive = 0000002C05DB8EB000000000, lsn = 5db8eb0/0\n"
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
"P00 INFO: check archive for segment 0000002C05DB8EB000000000\n"
Fix error in timeline conversion. The timeline is required to verify WAL segments in the archive after a backup. The conversion was performed base 10 instead of 16, which led to errors when the timeline was ≥ 0xA.
2020-01-21 10:29:46 -07:00
"P00 WARN: a timeline switch has occurred since the 20191027-181320F backup, enabling delta checksum\n"
Add hint when checksum delta is enabled after a timeline switch. This warning is normal when restoring a backup or promoting a standby so add a hint to make that clear.
2020-06-16 13:20:01 -04:00
" HINT: this is normal after restoring from backup or promoting a standby.\n"
Add timeline and checkpoint checks to backup. Add the following checks: * Checkpoint is updated in pg_control after pg_start_backup(). This helps ensure that PostgreSQL and pgBackRest have a consistent view of the storage and that PGDATA paths match. * Timeline of backup start WAL file matches pg_control. Hard to see how this one could get hit, but we have the power... * Standby is on the same timeline as the primary. If not, this standby is not following the primary. * Last standby checkpoint is not greater than the backup checkpoint. If so, this standby is not following the primary. This also requires some additional plumbing to read/write timeline/checkpoint from pg_control and parse timelines from WAL filenames. There were some changes in the backup tests caused by the fact that pg_control now has different contents for each backup. The check to ensure that the required checkpoint was reached on the standby should also be updated to use pg_control (it currently uses pg_control_checkpoint()), but that requires non-trivial changes to the test harness and will need to wait.
2021-12-07 09:21:07 -05:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/global/pg_control (8KB, [PCT]) checksum [SHA1]\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"P01 DETAIL: match file from prior backup " TEST_PATH "/pg1/base/1/1 (8KB, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: match file from prior backup " TEST_PATH "/pg1/postgresql.conf (11B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: match file from prior backup " TEST_PATH "/pg1/PG_VERSION (2B, [PCT]) checksum [SHA1]\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 DETAIL: hardlink pg_data/PG_VERSION to 20191027-181320F\n"
Add backup delta unit test. This test was added to take the place of another test, which turned out not to be workable. Even so, it adds coverages at little cost so it seems worth keeping.
2021-03-11 14:40:14 -05:00
"P00 DETAIL: hardlink pg_data/base/1/1 to 20191027-181320F\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 DETAIL: hardlink pg_data/postgresql.conf to 20191027-181320F\n"
"P00 DETAIL: hardlink pg_tblspc/32768/PG_11_201809051/1/5 to 20191027-181320F\n"
"P00 INFO: execute non-exclusive pg_stop_backup() and wait for all WAL segments to archive\n"
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
"P00 INFO: backup stop archive = 0000002C05DB8EB000000001, lsn = 5db8eb0/180000\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"P00 DETAIL: wrote 'backup_label' file returned from pg_stop_backup()\n"
Add tablespace_map file to command/backup test module. The code worked fine but better to have explicit tests for this file.
2021-11-15 14:32:22 -05:00
"P00 DETAIL: wrote 'tablespace_map' file returned from pg_stop_backup()\n"
Check archive immediately after backup start. Previously the archive was only checked at the end of the backup to ensure all WAL required to make the backup consistent was present. The problem was that if archiving was not functioning then the backup had to complete before the user found out, which could be a while if the database was large enough. Add an archive check immediately after backup start so failures are reported earlier. The trick is to determine which WAL to check. If the repo is new there may not be any WAL in it and pg_start_backup() will not switch the WAL segment if it is empty. These are both likely scenarios when setting up and/or testing pgBackRest. If the WAL segment is switched by pg_start_backup(), then check the archive for the segment that was detected prior to backup start. This should be common on normal running clusters with regular activity. Note that this might not be the segment immediately prior to the backup start segment if WAL volume is high. If pg_start_backup() did not switch the WAL then we can force a switch on PostgreSQL >= 9.3 by creating a restore point. In that case the WAL to check will be the backup start WAL. This is most likely to happen on idle systems, during testing, or immediately after a repo switch. An advantage of this approach other than earlier notification is that the backup directory will not be created so no resume will be attempted on the next backup. Note that some additional churn was created in backup.c because the load of archive.info needs to be done earlier.
2021-11-18 16:18:10 -05:00
"P00 INFO: check archive for segment(s) 0000002C05DB8EB000000000:0000002C05DB8EB000000001\n"
Log backup file total and restore size/file total. The backup size was a bit off because it did not include any files (e.g. backup_label, WAL files) that were added to the manifest after the main copy. To fix this move the log message to the very end of the backup. Add size/file total log message to restore since it did not exist before.
2021-08-11 13:39:36 -04:00
"P00 INFO: new backup label = 20191027-181320F_20191030-014640I\n"
Add tablespace_map file to command/backup test module. The code worked fine but better to have explicit tests for this file.
2021-11-15 14:32:22 -05:00
"P00 INFO: incr backup size = [SIZE], file total = 7");
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
Remove last vestiges of key replacement missed in b270253a.
2021-07-15 13:19:49 -04:00
TEST_RESULT_STR_Z(
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
testBackupValidate(storageRepo(), STRDEF(STORAGE_REPO_BACKUP "/latest")),
". {link, d=20191027-181320F_20191030-014640I}\n"
"pg_data {path}\n"
"pg_data/PG_VERSION.gz {file, s=2}\n"
"pg_data/backup_label.gz {file, s=17}\n"
"pg_data/base {path}\n"
Add backup delta unit test. This test was added to take the place of another test, which turned out not to be workable. Even so, it adds coverages at little cost so it seems worth keeping.
2021-03-11 14:40:14 -05:00
"pg_data/base/1 {path}\n"
"pg_data/base/1/1.gz {file, s=8192}\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"pg_data/global {path}\n"
"pg_data/global/pg_control.gz {file, s=8192}\n"
"pg_data/pg_tblspc {path}\n"
"pg_data/pg_tblspc/32768 {link, d=../../pg_tblspc/32768}\n"
"pg_data/pg_wal {path}\n"
"pg_data/postgresql.conf.gz {file, s=11}\n"
Add tablespace_map file to command/backup test module. The code worked fine but better to have explicit tests for this file.
2021-11-15 14:32:22 -05:00
"pg_data/tablespace_map.gz {file, s=19}\n"
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"pg_tblspc {path}\n"
"pg_tblspc/32768 {path}\n"
"pg_tblspc/32768/PG_11_201809051 {path}\n"
"pg_tblspc/32768/PG_11_201809051/1 {path}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"pg_tblspc/32768/PG_11_201809051/1/5.gz {file, s=0}\n"
"--------\n"
"[backup:target]\n"
Add defines for many test*() getter functions. A define was already added for TEST_PATH but it was not widely used. Replace all occurrences of testPath() with TEST_PATH in the tests. Replace testUser() with TEST_USER, testGroup() with TEST_GROUP, testRepoPath() with HRN_PATH_REPO, testDataPath() with HRN_PATH, testProjectExe() with TEST_PROJECT_EXE, and testScale() with TEST_SCALE. Replace {[path]}, {[user]}, {[group]}, etc. with defines and remove hrnReplaceKey(). This is better than having two ways to deal with replacements. In some cases the original test*() getters were kept because they are used by the harness, which does not have access to the new defines. Move them to harnessTest.intern.h to indicate that the tests should no longer use them.
2021-05-22 09:30:54 -04:00
"pg_data={\"path\":\"" TEST_PATH "/pg1\",\"type\":\"path\"}\n"
Fix spacing and typos in backup/manifest modules.
2020-11-09 16:26:43 -05:00
"pg_tblspc/32768={\"path\":\"../../pg1-tblspc/32768\",\"tablespace-id\":\"32768\""
",\"tablespace-name\":\"tblspc32768\",\"type\":\"link\"}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"\n"
"[target:file]\n"
"pg_data/PG_VERSION={\"checksum\":\"17ba0791499db908433b80f37c5fbc89b870084b\",\"reference\":\"20191027-181320F\""
",\"size\":2,\"timestamp\":1572200000}\n"
"pg_data/backup_label={\"checksum\":\"8e6f41ac87a7514be96260d65bacbffb11be77dc\",\"size\":17"
",\"timestamp\":1572400002}\n"
Add backup delta unit test. This test was added to take the place of another test, which turned out not to be workable. Even so, it adds coverages at little cost so it seems worth keeping.
2021-03-11 14:40:14 -05:00
"pg_data/base/1/1={\"checksum\":\"0631457264ff7f8d5fb1edc2c0211992a67c73e6\",\"checksum-page\":true"
Remove primary flag from manifest. This flag was only being used by the backup command after manifestNewBuild() and had no other uses. There was a time when it was important for integration testing but the unit tests now fulfill this role. Since backup is the only code concerned with the primary flag, move the code into the backup module. We don't have any cross-version testing but this change was tested manually with the most recent version of pgBackRest to make sure it was tolerant of the missing primary info. When an older version of pgBackRest loads a newer manifest the primary flag will always be set to false, which is fine since it is not used.
2022-01-20 14:01:10 -05:00
",\"reference\":\"20191027-181320F\",\"size\":8192,\"timestamp\":1572200000}\n"
Add timeline and checkpoint checks to backup. Add the following checks: * Checkpoint is updated in pg_control after pg_start_backup(). This helps ensure that PostgreSQL and pgBackRest have a consistent view of the storage and that PGDATA paths match. * Timeline of backup start WAL file matches pg_control. Hard to see how this one could get hit, but we have the power... * Standby is on the same timeline as the primary. If not, this standby is not following the primary. * Last standby checkpoint is not greater than the backup checkpoint. If so, this standby is not following the primary. This also requires some additional plumbing to read/write timeline/checkpoint from pg_control and parse timelines from WAL filenames. There were some changes in the backup tests caused by the fact that pg_control now has different contents for each backup. The check to ensure that the required checkpoint was reached on the standby should also be updated to use pg_control (it currently uses pg_control_checkpoint()), but that requires non-trivial changes to the test harness and will need to wait.
2021-12-07 09:21:07 -05:00
"pg_data/global/pg_control={\"size\":8192,\"timestamp\":1572400000}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"pg_data/postgresql.conf={\"checksum\":\"e3db315c260e79211b7b52587123b7aa060f30ab\""
",\"reference\":\"20191027-181320F\",\"size\":11,\"timestamp\":1570000000}\n"
Add tablespace_map file to command/backup test module. The code worked fine but better to have explicit tests for this file.
2021-11-15 14:32:22 -05:00
"pg_data/tablespace_map={\"checksum\":\"87fe624d7976c2144e10afcb7a9a49b071f35e9c\",\"size\":19"
",\"timestamp\":1572400002}\n"
Remove primary flag from manifest. This flag was only being used by the backup command after manifestNewBuild() and had no other uses. There was a time when it was important for integration testing but the unit tests now fulfill this role. Since backup is the only code concerned with the primary flag, move the code into the backup module. We don't have any cross-version testing but this change was tested manually with the most recent version of pgBackRest to make sure it was tolerant of the missing primary info. When an older version of pgBackRest loads a newer manifest the primary flag will always be set to false, which is fine since it is not used.
2022-01-20 14:01:10 -05:00
"pg_tblspc/32768/PG_11_201809051/1/5={\"checksum-page\":true,\"reference\":\"20191027-181320F\",\"size\":0"
",\"timestamp\":1572200000}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"\n"
"[target:link]\n"
"pg_data/pg_tblspc/32768={\"destination\":\"../../pg1-tblspc/32768\"}\n"
"\n"
"[target:path]\n"
"pg_data={}\n"
"pg_data/base={}\n"
Add backup delta unit test. This test was added to take the place of another test, which turned out not to be workable. Even so, it adds coverages at little cost so it seems worth keeping.
2021-03-11 14:40:14 -05:00
"pg_data/base/1={}\n"
Added redacted manifest to testBackupValidate(). The manifest is excellent for validation but including the entire manifest is too noisy and some values are architecture/algorithm dependent. Output a redacted version that contains the most important information which can be improved on over time.
2020-03-18 10:10:10 -04:00
"pg_data/global={}\n"
"pg_data/pg_tblspc={}\n"
"pg_data/pg_wal={}\n"
"pg_tblspc={}\n"
"pg_tblspc/32768={}\n"
"pg_tblspc/32768/PG_11_201809051={}\n"
"pg_tblspc/32768/PG_11_201809051/1={}\n",
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
"compare file list");
}
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
// -------------------------------------------------------------------------------------------------------------------------
TEST_TITLE("online 11 full backup with tablespaces and bundles");
backupTimeStart = BACKUP_EPOCH + 2400000;
{
// Load options
StringList *argList = strLstNew();
hrnCfgArgRawZ(argList, cfgOptStanza, "test1");
hrnCfgArgRaw(argList, cfgOptRepoPath, repoPath);
hrnCfgArgRaw(argList, cfgOptPgPath, pg1Path);
hrnCfgArgRawZ(argList, cfgOptRepoRetentionFull, "1");
hrnCfgArgRawStrId(argList, cfgOptType, backupTypeFull);
hrnCfgArgRawZ(argList, cfgOptManifestSaveThreshold, "1");
hrnCfgArgRawBool(argList, cfgOptArchiveCopy, true);
hrnCfgArgRawZ(argList, cfgOptBufferSize, "16K");
Rename bundle-* options to repo-bundle-*. It seems best for these to be repo options so they can be configured per repo, rather than globally. All clarify usage for repo-bundle-size and repo-bundle-limit.
2022-03-14 17:49:52 -06:00
hrnCfgArgRawBool(argList, cfgOptRepoBundle, true);
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
HRN_CFG_LOAD(cfgCmdBackup, argList);
// Set to a smaller values than the defaults allow
Rename bundle-* options to repo-bundle-*. It seems best for these to be repo options so they can be configured per repo, rather than globally. All clarify usage for repo-bundle-size and repo-bundle-limit.
2022-03-14 17:49:52 -06:00
cfgOptionSet(cfgOptRepoBundleSize, cfgSourceParam, VARINT64(PG_PAGE_SIZE_DEFAULT));
cfgOptionSet(cfgOptRepoBundleLimit, cfgSourceParam, VARINT64(PG_PAGE_SIZE_DEFAULT));
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
// Zeroed file which passes page checksums
Buffer *relation = bufNew(PG_PAGE_SIZE_DEFAULT * 3);
memset(bufPtr(relation), 0, bufSize(relation));
bufUsedSet(relation, bufSize(relation));
HRN_STORAGE_PUT(storagePgWrite(), PG_PATH_BASE "/1/2", relation, .timeModified = backupTimeStart);
// Old files
HRN_STORAGE_PUT_Z(storagePgWrite(), "postgresql.auto.conf", "CONFIGSTUFF2", .timeModified = 1500000000);
HRN_STORAGE_PUT_Z(storagePgWrite(), "stuff.conf", "CONFIGSTUFF3", .timeModified = 1500000000);
// File that will get skipped while bundling smaller files and end up a bundle by itself
Buffer *bigish = bufNew(PG_PAGE_SIZE_DEFAULT - 1);
memset(bufPtr(bigish), 0, bufSize(bigish));
bufUsedSet(bigish, bufSize(bigish));
HRN_STORAGE_PUT(storagePgWrite(), "bigish.dat", bigish, .timeModified = 1500000001);
// Run backup
testBackupPqScriptP(PG_VERSION_11, backupTimeStart, .walCompressType = compressTypeGz, .walTotal = 2);
Show backup percent complete in info output. In the JSON output the percent complete is storage as an integer of the percent complete * 100. So, before display it should be converted to double and divided by 100, or split using integer mod and div. Note that percent complete will only be displayed on the host where the backup was executed. Remote hosts will show a backup/expire running with no percent complete.
2022-05-04 12:52:05 -04:00
TEST_RESULT_VOID(testCmdBackup(), "backup");
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
TEST_RESULT_LOG(
"P00 INFO: execute non-exclusive pg_start_backup(): backup begins after the next regular checkpoint completes\n"
"P00 INFO: backup start archive = 0000000105DB8EB000000000, lsn = 5db8eb0/0\n"
"P00 INFO: check archive for segment 0000000105DB8EB000000000\n"
"P00 DETAIL: store zero-length file " TEST_PATH "/pg1/pg_tblspc/32768/PG_11_201809051/1/5\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/base/1/2 (24KB, [PCT]) checksum [SHA1]\n"
Add bundle logging to backup command. This was added to the restore command so add it to the backup command as well.
2022-03-09 15:34:15 -06:00
"P01 DETAIL: backup file " TEST_PATH "/pg1/stuff.conf (bundle 1/0, 12B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/postgresql.auto.conf (bundle 1/32, 12B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/postgresql.conf (bundle 1/64, 11B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/PG_VERSION (bundle 1/95, 2B, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/bigish.dat (bundle 2/0, 8.0KB, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/base/1/1 (bundle 3/0, 8KB, [PCT]) checksum [SHA1]\n"
"P01 DETAIL: backup file " TEST_PATH "/pg1/global/pg_control (bundle 4/0, 8KB, [PCT]) checksum [SHA1]\n"
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
"P00 INFO: execute non-exclusive pg_stop_backup() and wait for all WAL segments to archive\n"
"P00 INFO: backup stop archive = 0000000105DB8EB000000001, lsn = 5db8eb0/180000\n"
"P00 DETAIL: wrote 'backup_label' file returned from pg_stop_backup()\n"
"P00 DETAIL: wrote 'tablespace_map' file returned from pg_stop_backup()\n"
"P00 INFO: check archive for segment(s) 0000000105DB8EB000000000:0000000105DB8EB000000001\n"
"P00 DETAIL: copy segment 0000000105DB8EB000000000 to backup\n"
"P00 DETAIL: copy segment 0000000105DB8EB000000001 to backup\n"
"P00 INFO: new backup label = 20191030-014640F\n"
"P00 INFO: full backup size = [SIZE], file total = 13");
TEST_RESULT_STR_Z(
testBackupValidate(storageRepo(), STRDEF(STORAGE_REPO_BACKUP "/latest")),
". {link, d=20191030-014640F}\n"
"bundle {path}\n"
Limit files that can be bundled. Limit which files can be added to bundles, which allows resume to work reasonably well. On resume, the bundles are removed and any remaining file is eligible to be to be resumed. Also reduce the bundle-size default to 20MiB. This is pretty arbitrary, but a smaller default seems better.
2022-02-17 07:25:12 -06:00
"bundle/1/pg_data/PG_VERSION {file, s=2}\n"
"bundle/1/pg_data/postgresql.auto.conf {file, s=12}\n"
"bundle/1/pg_data/postgresql.conf {file, s=11}\n"
"bundle/1/pg_data/stuff.conf {file, s=12}\n"
"bundle/2/pg_data/bigish.dat {file, s=8191}\n"
"bundle/3/pg_data/base/1/1 {file, s=8192}\n"
"bundle/4/pg_data/global/pg_control {file, s=8192}\n"
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
"pg_data {path}\n"
"pg_data/backup_label.gz {file, s=17}\n"
Limit files that can be bundled. Limit which files can be added to bundles, which allows resume to work reasonably well. On resume, the bundles are removed and any remaining file is eligible to be to be resumed. Also reduce the bundle-size default to 20MiB. This is pretty arbitrary, but a smaller default seems better.
2022-02-17 07:25:12 -06:00
"pg_data/base {path}\n"
"pg_data/base/1 {path}\n"
"pg_data/base/1/2.gz {file, s=24576}\n"
Bundle files in the repository during backup. Bundle (combine) smaller files during backup to reduce the number of files written to the repository (enable with --bundle). Reducing the number of files is a benefit on all file systems, but especially so on object stores such as S3 that have a high file creation cost. Another benefit is that zero-length files are only stored as metadata in the manifest. Files are batched up to bundle-size and then compressed/encrypted individually and stored sequentially in the bundle. The bundle id and offset of each file is stored in the manifest so files can be retrieved randomly without needing to read the entire bundle. Files are ordered by timestamp descending when being assigned to bundles to reduce the amount of random access that needs to be done. The idea is that bundles with older files can be read in their entirety on restore and only bundles with newer files will get fragmented. Bundles are a custom format with metadata stored in the manifest. Tar was considered but it is too limited a format, the major issue being that the size of the file must be known in advance and that is very contrary to how pgBackRest works, especially once we introduce page-level incremental backups. Bundles are stored numbered in the bundle directory. Some files may still end up in pg_data if they are added after the backup is complete. backup_label is an example. Currently, only the backup command works in batches. The restore and verify commands use the offsets to pull individual files out of the bundle. It seems better to finalize how this is going to work before optimizing the other commands. Even as is, this is a major step forward, and all commands function with bundling. One caveat: resume is currently not supported when bundle is enabled.
2022-02-14 13:24:14 -06:00
"pg_data/pg_wal {path}\n"
"pg_data/pg_wal/0000000105DB8EB000000000.gz {file, s=1048576}\n"
"pg_data/pg_wal/0000000105DB8EB000000001.gz {file, s=1048576}\n"
"pg_data/tablespace_map.gz {file, s=19}\n"
"--------\n"
"[backup:target]\n"
"pg_data={\"path\":\"" TEST_PATH "/pg1\",\"type\":\"path\"}\n"
"pg_tblspc/32768={\"path\":\"../../pg1-tblspc/32768\",\"tablespace-id\":\"32768\""
",\"tablespace-name\":\"tblspc32768\",\"type\":\"link\"}\n"
"\n"
"[target:file]\n"
"pg_data/PG_VERSION={\"checksum\":\"17ba0791499db908433b80f37c5fbc89b870084b\",\"size\":2"
",\"timestamp\":1572200000}\n"
"pg_data/backup_label={\"checksum\":\"8e6f41ac87a7514be96260d65bacbffb11be77dc\",\"size\":17"
",\"timestamp\":1572400002}\n"
"pg_data/base/1/1={\"checksum\":\"0631457264ff7f8d5fb1edc2c0211992a67c73e6\",\"checksum-page\":true,\"size\":8192"
",\"timestamp\":1572200000}\n"
"pg_data/base/1/2={\"checksum\":\"ebdd38b69cd5b9f2d00d273c981e16960fbbb4f7\",\"checksum-page\":true,\"size\":24576"
",\"timestamp\":1572400000}\n"
"pg_data/bigish.dat={\"checksum\":\"3e5175386be683d2f231f3fa3eab892a799082f7\",\"size\":8191"
",\"timestamp\":1500000001}\n"
"pg_data/global/pg_control={\"size\":8192,\"timestamp\":1572400000}\n"
"pg_data/pg_wal/0000000105DB8EB000000000={\"size\":1048576,\"timestamp\":1572400002}\n"
"pg_data/pg_wal/0000000105DB8EB000000001={\"size\":1048576,\"timestamp\":1572400002}\n"
"pg_data/postgresql.auto.conf={\"checksum\":\"e873a5cb5a67e48761e7b619c531311404facdce\",\"size\":12"
",\"timestamp\":1500000000}\n"
"pg_data/postgresql.conf={\"checksum\":\"e3db315c260e79211b7b52587123b7aa060f30ab\",\"size\":11"
",\"timestamp\":1570000000}\n"
"pg_data/stuff.conf={\"checksum\":\"55a9d0d18b77789c7722abe72aa905e2dc85bb5d\",\"size\":12"
",\"timestamp\":1500000000}\n"
"pg_data/tablespace_map={\"checksum\":\"87fe624d7976c2144e10afcb7a9a49b071f35e9c\",\"size\":19"
",\"timestamp\":1572400002}\n"
"pg_tblspc/32768/PG_11_201809051/1/5={\"checksum-page\":true,\"size\":0,\"timestamp\":1572200000}\n"
"\n"
"[target:link]\n"
"pg_data/pg_tblspc/32768={\"destination\":\"../../pg1-tblspc/32768\"}\n"
"\n"
"[target:path]\n"
"pg_data={}\n"
"pg_data/base={}\n"
"pg_data/base/1={}\n"
"pg_data/global={}\n"
"pg_data/pg_tblspc={}\n"
"pg_data/pg_wal={}\n"
"pg_tblspc={}\n"
"pg_tblspc/32768={}\n"
"pg_tblspc/32768/PG_11_201809051={}\n"
"pg_tblspc/32768/PG_11_201809051/1={}\n",
"compare file list");
}
The backup command is implemented entirely in C. For the most part this is a direct migration of the Perl code into C except as noted below. A backup can now be initiated from a linked directory. The link will not be stored in the manifest or recreated on restore. If a link or directory does not already exist in the restore location then a directory will be created. The logic for creating backup labels has been improved and it should no longer be possible to get a backup label earlier than the latest backup even with timezone changes or clock skew. This has never been an issue in the field that we know of, but we found it in testing. For online backups all times are fetched from the PostgreSQL primary host (before only copy start was). This doesn't affect backup integrity but it does prevent clock skew between hosts affecting backup duration reporting. Archive copy now works as expected when the archive and backup have different compression settings, i.e. when one is compressed and the other is not. This was a long-standing bug in the Perl code. Resume will now work even if hardlink settings have been changed. Reviewed by Cynthia Shang.
2019-12-13 17:14:26 -05:00
}
Rename FUNCTION_HARNESS_RESULT*() macros to FUNCTION_HARNESS_RETURN*(). When the FUNCTION_*_RESULT*() macros were renamed to FUNCTION_*_RETURN_*() in the core code the test harness macros were missed. Update them to make the naming consistent.
2021-03-10 18:42:22 -05:00
FUNCTION_HARNESS_RETURN_VOID();
The local command for backup is implemented entirely in C. The local process is now entirely migrated to C. Since all major I/O operations are performed in the local process, the vast majority of I/O is now performed in C. Contributed by David Steele, Cynthia Shang.
2019-07-25 14:34:16 -04:00
}
Reference in New Issue Copy Permalink