The names were changed in PostgreSQL 15, so update the code and docs to make the naming more generic where needed to avoid using a version-specific name in the logs and documentation.
The prior result was hex-encoded, which is not optimal. This was legacy from the interface with Perl and then the JSON protocol. The new binary protocol natively supports binary so it makes sense to use it and convert to hex where needed.
A number of these hex conversions can now be removed but that will need to be handled in another commit.
The reference list was previously built at load time from whichever references existed in the file list. This was sufficient since the list was for informational purposes only.
The block incremental feature will require a reference list that contains all prior backups, even those that are not explicitly referenced from the manifest. Therefore it makes sense to build and persist a manifest list rather than building it at load time.
This list can still be used for informational purposes, though it needs to be sorted since the list it sill built for older manifest versions and may not be in sorted order.
Add strLstFindIdx() to find references in the list.
The prior method was to check a combination of fields to determine if a file needed to be copied, delta'd, or resumed. This was complicated and ultimately imposed a limitation on the number of operations that could be performed.
Introduce copy, delta, and resume flags in the manifest to make it clearer which operations need to be performed and to reduce complex and duplicated logic.
This also allows zero-length bundled files to be completed during manifest build rather than later on during backup processing.
This appears to have been an oversight in 34d6495. Storing the reference is not really correct since the file is not stored in a prior backup. It also uses more space.
There is no real harm in storing the reference, since it is always ignored on restore, but the code is simpler if the zero-length files can be dealt with during the manifest and don't need additional handling later on. This is also an important part of some upcoming optimizations.
It is possible to log the bundle info correctly but the information is useless with the backup reference, which does not appear until later. For now just omit the bundle info so we are not logging something incorrect.
This test exposes a small logging issue. The bundle information for the matched delta on PG_VERSION is not correct. This issue will be fixed in the next commit.
The information stored in the manifest *is* correct so this bug is essentially cosmetic.
Lifecycle policies can cause the archive.info file and its copy to be removed since they are only updated on a stanza-upgrade. Update the timestamps after a successful backup to prevent this.
This does not mean that lifecycle policies should be used as a replacement for expiration. However, in some cases there may be policies in place that are out of admin control. If the lifecycle expiration is less than pgbackrest expiration then corruption of the earliest backup will occur at the very least and there might be other corruption which would make the repo unrecoverable.
Allow key/value annotations to be added with the backup command and added/modified/removed with the new annotate command.
Annotations can be viewed with the info command in text mode when --set is specified and are always included in JSON output.
The prior warning made it sound as if some action was required on the part of the user.
The new message should make it clearer that this action will be performed by pgBackRest.
Previously a callback was used to list path contents and if no sort was specified then a snapshot was not required. When deleting files from the path some filesystems could omit files that still existed, which meant the path could not be removed.
Filter . out of lists in the Posix driver since this special entry was only used by test code (and filtered everywhere in the core code).
Also remove callbacks from the storage interface and replace with an iterator that should be easier to use and guarantees efficient use of the snapshots.
The manifest is saved on a regular basis during a backup so a failed backup can be resumed. For backups that the user has configured/invoked as not resumable, skip the incremental save of the manifest.
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.
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.
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'
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
Eliminate summing and passing of copied files sizes for logging backup size.
Instead, utilize infoBackupDataByLabel() to pull the backup size for the log message.
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.
