1
0
mirror of https://github.com/pgbackrest/pgbackrest.git synced 2024-12-14 10:13:05 +02:00
pgbackrest/lib/pgBackRest/Backup.pm

1081 lines
44 KiB
Perl
Raw Normal View History

v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
####################################################################################################################################
# BACKUP MODULE
####################################################################################################################################
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
package pgBackRest::Backup;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
use threads;
use strict;
use warnings FATAL => qw(all);
use Carp qw(confess);
2014-07-28 01:13:23 +03:00
use Exporter qw(import);
use Fcntl 'SEEK_CUR';
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
use File::Basename;
use File::Path qw(remove_tree);
use Thread::Queue;
use lib dirname($0);
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
use pgBackRest::Common::Exception;
use pgBackRest::Common::Exit;
use pgBackRest::Common::Ini;
use pgBackRest::Common::Log;
use pgBackRest::Archive;
use pgBackRest::BackupCommon;
use pgBackRest::BackupFile;
use pgBackRest::BackupInfo;
use pgBackRest::Common::String;
use pgBackRest::Config::Config;
use pgBackRest::Db;
use pgBackRest::File;
use pgBackRest::FileCommon;
use pgBackRest::Manifest;
use pgBackRest::Version;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
####################################################################################################################################
# Operation constants
####################################################################################################################################
use constant OP_BACKUP => 'Backup';
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
use constant OP_BACKUP_DESTROY => OP_BACKUP . '->DESTROY';
use constant OP_BACKUP_FILE_NOT_IN_MANIFEST => OP_BACKUP . '->fileNotInManifest';
use constant OP_BACKUP_NEW => OP_BACKUP . '->new';
use constant OP_BACKUP_PROCESS => OP_BACKUP . '->process';
use constant OP_BACKUP_PROCESS_MANIFEST => OP_BACKUP . '->processManifest';
use constant OP_BACKUP_TMP_CLEAN => OP_BACKUP . '->tmpClean';
use constant OP_BACKUP_TYPE_FIND => OP_BACKUP . '->typeFind';
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
####################################################################################################################################
# new
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
####################################################################################################################################
sub new
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
my $class = shift; # Class name
# Create the class hash
my $self = {};
bless $self, $class;
# Assign function parameters, defaults, and log debug info
my ($strOperation) = logDebugParam(OP_BACKUP_NEW);
# Initialize default file object
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
$self->{oFile} = new pgBackRest::File
(
optionGet(OPTION_STANZA),
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
optionGet(OPTION_REPO_PATH),
optionRemoteType(),
protocolGet()
);
# Initialize variables
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
$self->{oDb} = new pgBackRest::Db();
# Return from function and log return values if any
return logDebugReturn
(
$strOperation,
{name => 'self', value => $self}
);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
####################################################################################################################################
# DESTROY
####################################################################################################################################
sub DESTROY
{
my $self = shift;
# Assign function parameters, defaults, and log debug info
my
(
$strOperation
) =
logDebugParam
(
OP_BACKUP_DESTROY
);
undef($self->{oFile});
undef($self->{oDb});
# Return from function and log return values if any
return logDebugReturn
(
$strOperation
);
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
####################################################################################################################################
# typeFind
#
# Find the last backup depending on the type.
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
####################################################################################################################################
sub typeFind
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
my $self = shift;
# Assign function parameters, defaults, and log debug info
my
(
$strOperation,
$strType,
$strBackupClusterPath
) =
logDebugParam
(
OP_BACKUP_TYPE_FIND, \@_,
{name => 'strType'},
{name => 'strBackupClusterPath'}
);
my $strLabel;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
if ($strType eq BACKUP_TYPE_INCR)
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
$strLabel = ($self->{oFile}->list(PATH_BACKUP_CLUSTER, undef, backupRegExpGet(true, true, true), 'reverse'))[0];
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
if (!defined($strLabel) && $strType ne BACKUP_TYPE_FULL)
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
$strLabel = ($self->{oFile}->list(PATH_BACKUP_CLUSTER, undef, backupRegExpGet(true), 'reverse'))[0];
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
2014-06-04 05:02:56 +03:00
# Return from function and log return values if any
return logDebugReturn
(
$strOperation,
{name => 'strLabel', value => $strLabel}
);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
####################################################################################################################################
# fileNotInManifest
#
# Find all files in a backup path that are not in the supplied manifest.
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
####################################################################################################################################
sub fileNotInManifest
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
my $self = shift;
# Assign function parameters, defaults, and log debug info
my
(
$strOperation,
$strPathType,
$oManifest,
$oAbortedManifest
) =
logDebugParam
(
OP_BACKUP_FILE_NOT_IN_MANIFEST, \@_,
{name => 'strPathType', trace => true},
{name => 'oManifest', trace => true},
{name => 'oAbortedManifest', trace => true}
);
# Build manifest for aborted temp path
2014-09-14 22:31:56 +03:00
my %oFileHash;
$self->{oFile}->manifest($strPathType, undef, \%oFileHash);
# Get compress flag
my $bCompressed = $oAbortedManifest->boolGet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_COMPRESS);
2014-09-14 22:31:56 +03:00
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
my @stryFile;
foreach my $strName (sort(keys(%{$oFileHash{name}})))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
# Ignore certain files that will never be in the manifest
if ($strName eq FILE_MANIFEST ||
$strName eq '.')
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
next;
}
# We'll always keep the base path
if ($strName eq MANIFEST_KEY_BASE)
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
next;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
# Keep the tablespace path if some tablespaces exist in the new manfest
elsif ($strName eq MANIFEST_TABLESPACE)
{
my $bFound = false;
foreach my $strPath ($oManifest->keys(MANIFEST_SECTION_BACKUP_PATH))
{
if ($strPath =~ /^$strName\//)
{
$bFound = true;
last;
}
}
next if $bFound;
}
# If there is a / in the name then check further, otherwise it's a temp file or some other garbage and should be deleted
elsif (index($strName, '/') != -1)
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
my $strBasePath = (split('/', $strName))[0];
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
my $strPath = substr($strName, length($strBasePath) + 1);
# Create the section from the base path
my $strSection = $strBasePath;
# Test to see if a tablespace exists in the new manifest
if ($strSection eq 'tablespace')
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
my $strTablespace = (split('/', $strPath))[0];
my $strTablespacePath = (split('/', $strPath))[1];
if (defined($strTablespacePath) && $strTablespacePath ne $oManifest->tablespacePathGet())
{
next;
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
$strSection = $strSection . '/' . $strTablespace;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
if (($strTablespace eq $strPath) || "${strTablespace}/${strTablespacePath}" eq $strPath)
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
if ($oManifest->test("${strSection}:path"))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
next;
}
}
$strPath = substr($strPath, length($strTablespace) + length($oManifest->tablespacePathGet()) + 2);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
# Get the file type (all links will be deleted since they are easy to recreate)
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
my $cType = $oFileHash{name}{"${strName}"}{type};
# If a directory check if it exists in the new manifest
if ($cType eq 'd')
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
if ($oManifest->test("${strSection}:path", "${strPath}"))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
next;
}
}
# Else if a file
elsif ($cType eq 'f')
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
# If the original backup was compressed the remove the extension before checking the manifest
if ($bCompressed)
{
$strPath = substr($strPath, 0, length($strPath) - 3);
}
# To be preserved the file must exist in the new manifest and not be a reference to a previous backup
if ($oManifest->test("${strSection}:file", $strPath) &&
!$oManifest->test("${strSection}:file", $strPath, MANIFEST_SUBKEY_REFERENCE))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
# To be preserved the checksum must be defined
my $strChecksum = $oAbortedManifest->get("${strSection}:file", $strPath, MANIFEST_SUBKEY_CHECKSUM, false);
# The timestamp should also match and the size if the file is not compressed. If the file is compressed it's
# not worth extracting the size - it will be hashed later to verify its authenticity.
if (defined($strChecksum) &&
($bCompressed || ($oManifest->numericGet("${strSection}:file", $strPath, MANIFEST_SUBKEY_SIZE) ==
$oFileHash{name}{$strName}{size})) &&
$oManifest->numericGet("${strSection}:file", $strPath, MANIFEST_SUBKEY_TIMESTAMP) ==
$oFileHash{name}{$strName}{modification_time})
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
$oManifest->set("${strSection}:file", $strPath, MANIFEST_SUBKEY_CHECKSUM, $strChecksum);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
next;
}
}
}
}
# Push the file/path/link to be deleted into the result array
push @stryFile, $strName;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
# Return from function and log return values if any
return logDebugReturn
(
$strOperation,
{name => 'stryFile', value => \@stryFile}
);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
####################################################################################################################################
# tmpClean
2014-06-04 18:58:30 +03:00
#
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Cleans the temp directory from a previous failed backup so it can be reused
####################################################################################################################################
sub tmpClean
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
my $self = shift;
# Assign function parameters, defaults, and log debug info
my
(
$strOperation,
$oManifest,
$oAbortedManifest
) =
logDebugParam
(
OP_BACKUP_TMP_CLEAN, \@_,
{name => 'oManifest', trace => true},
{name => 'oAbortedManifest', trace => true}
);
&log(INFO, 'clean backup temp path: ' . $self->{oFile}->pathGet(PATH_BACKUP_TMP));
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Remove the pg_xlog directory since it contains nothing useful for the new backup
if (-e $self->{oFile}->pathGet(PATH_BACKUP_TMP, 'base/pg_xlog'))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
remove_tree($self->{oFile}->pathGet(PATH_BACKUP_TMP, 'base/pg_xlog'))
or confess &log(ERROR, 'unable to delete tmp pg_xlog path');
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
# Remove the pg_tblspc directory since it is trivial to rebuild, but hard to compare
if (-e $self->{oFile}->pathGet(PATH_BACKUP_TMP, 'base/pg_tblspc'))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
remove_tree($self->{oFile}->pathGet(PATH_BACKUP_TMP, 'base/pg_tblspc'))
or confess &log(ERROR, 'unable to delete tmp pg_tblspc path');
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
# Get the list of files that should be deleted from temp
my @stryFile = $self->fileNotInManifest(PATH_BACKUP_TMP, $oManifest, $oAbortedManifest);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
foreach my $strFile (sort {$b cmp $a} @stryFile)
{
my $strDelete = $self->{oFile}->pathGet(PATH_BACKUP_TMP, $strFile);
2014-06-04 18:58:30 +03:00
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# If a path then delete it, all the files should have already been deleted since we are going in reverse order
if (-d $strDelete)
{
logDebugMisc($strOperation, "remove path ${strDelete}");
rmdir($strDelete)
or confess &log(ERROR, "unable to delete path ${strDelete}, is it empty?", ERROR_PATH_REMOVE);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
# Else delete a file
else
{
logDebugMisc($strOperation, "remove file ${strDelete}");
fileRemove($strDelete);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
}
# Return from function and log return values if any
return logDebugReturn
(
$strOperation
);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
####################################################################################################################################
# processManifest
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
#
# Process the file level backup. Uses the information in the manifest to determine which files need to be copied. Directories
# and tablespace links are only created when needed, except in the case of a full backup or if hardlinks are requested.
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
####################################################################################################################################
sub processManifest
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
my $self = shift;
# Assign function parameters, defaults, and log debug info
my
(
$strOperation,
$strType,
$bCompress,
$bHardLink,
$oBackupManifest # Manifest for the current backup
) =
logDebugParam
(
OP_BACKUP_PROCESS_MANIFEST, \@_,
{name => 'strType'},
{name => 'bCompress'},
{name => 'bHardLink'},
{name => 'oBackupManifest'},
);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Variables used for parallel copy
my %oFileCopyMap;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
my $lFileTotal = 0;
my $lSizeTotal = 0;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Determine whether all paths and links will be created
my $bFullCreate = $bHardLink || $strType eq BACKUP_TYPE_FULL;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Iterate through the path sections of the manifest to backup
foreach my $strPathKey ($oBackupManifest->keys(MANIFEST_SECTION_BACKUP_PATH))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
# Determine the source and destination backup paths
my $strBackupSourcePath; # Absolute path to the database base directory or tablespace to backup
my $strBackupDestinationPath; # Relative path to the backup directory where the data will be stored
$strBackupSourcePath = $oBackupManifest->get(MANIFEST_SECTION_BACKUP_PATH, $strPathKey, MANIFEST_SUBKEY_PATH);
$strBackupDestinationPath = $strPathKey;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Create links for tablespaces
if ($oBackupManifest->test(MANIFEST_SECTION_BACKUP_PATH, $strPathKey, MANIFEST_SUBKEY_LINK))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
if ($bFullCreate)
{
$self->{oFile}->linkCreate(PATH_BACKUP_TMP, $strBackupDestinationPath,
PATH_BACKUP_TMP,
'base/pg_tblspc/' . $oBackupManifest->get(MANIFEST_SECTION_BACKUP_PATH,
$strPathKey, MANIFEST_SUBKEY_LINK),
false, true, true);
}
if ($oBackupManifest->numericGet(MANIFEST_SECTION_BACKUP_DB, MANIFEST_KEY_DB_VERSION) >= 9.0)
{
$strBackupSourcePath .= '/' . $oBackupManifest->tablespacePathGet();
$strBackupDestinationPath .= '/' . $oBackupManifest->tablespacePathGet();
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
# If this is a full backup or hard-linked then create all paths and links
if ($bFullCreate)
{
# Create paths
my $strSectionPath = "$strPathKey:path";
if ($oBackupManifest->test($strSectionPath))
{
foreach my $strPath ($oBackupManifest->keys($strSectionPath))
{
if ($strPath ne '.')
{
$self->{oFile}->pathCreate(PATH_BACKUP_TMP, "${strBackupDestinationPath}/${strPath}");
}
}
}
# Create links
#
# Non-tablespace links are no longer created in backup directories because they are potentially dangerous.
# This feature may be brought back at a later date but more likely that it will be rethought completely.
#
# my $strSectionLink = "$strPathKey:link";
#
# if ($oBackupManifest->test($strSectionLink))
# {
# foreach my $strLink ($oBackupManifest->keys($strSectionLink))
# {
# # Create links except in pg_tblspc because they have already been created
# if (!($strPathKey eq 'base' && $strLink =~ /^pg_tblspc\/.*/))
# {
# $self->{oFile}->linkCreate(PATH_BACKUP_ABSOLUTE,
# $oBackupManifest->get($strSectionLink, $strLink, MANIFEST_SUBKEY_DESTINATION),
# PATH_BACKUP_TMP, "${strBackupDestinationPath}/${strLink}",
# false, false, false);
# }
# }
# }
}
# Possible for the file section to exist with no files (i.e. empty tablespace)
my $strSectionFile = "$strPathKey:file";
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Iterate through the files for each backup source path
2015-01-21 17:44:08 +02:00
foreach my $strFile ($oBackupManifest->keys($strSectionFile))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
my $strBackupSourceFile = "${strBackupSourcePath}/${strFile}";
2014-06-04 18:58:30 +03:00
# If the file has a reference it does not need to be copied since it can be retrieved from the referenced backup.
# However, if hard-linking is turned on the link will need to be created
my $bProcess = true;
my $strReference = $oBackupManifest->get($strSectionFile, $strFile, MANIFEST_SUBKEY_REFERENCE, false);
if (defined($strReference))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
# If hardlinking is turned on then create a hardlink for files that have not changed since the last backup
if ($bHardLink)
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
logDebugMisc($strOperation, "hardlink ${strBackupSourceFile} to ${strReference}");
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
$self->{oFile}->linkCreate(PATH_BACKUP_CLUSTER, "${strReference}/${strBackupDestinationPath}/${strFile}",
PATH_BACKUP_TMP, "${strBackupDestinationPath}/${strFile}", true, false, true);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
else
{
logDebugMisc($strOperation, "reference ${strBackupSourceFile} to ${strReference}");
}
$bProcess = false;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
if ($bProcess)
{
my $lFileSize = $oBackupManifest->numericGet($strSectionFile, $strFile, MANIFEST_SUBKEY_SIZE);
# Increment file total
$lFileTotal++;
my $strFileKey;
# Certain files are not copied until the end
if ($strPathKey eq MANIFEST_KEY_BASE && $strFile eq FILE_PG_CONTROL)
{
$strFileKey = $strFile;
$oFileCopyMap{$strPathKey}{$strFileKey}{skip} = true;
}
# Else continue normally
else
{
$strFileKey = sprintf("%016d-${strFile}", $lFileSize);
$oFileCopyMap{$strPathKey}{$strFileKey}{skip} = false;
# Add file size to total size
$lSizeTotal += $lFileSize;
}
$oFileCopyMap{$strPathKey}{$strFileKey}{db_file} = $strBackupSourceFile;
$oFileCopyMap{$strPathKey}{$strFileKey}{file_section} = $strSectionFile;
$oFileCopyMap{$strPathKey}{$strFileKey}{file} = ${strFile};
$oFileCopyMap{$strPathKey}{$strFileKey}{backup_file} = "${strBackupDestinationPath}/${strFile}";
$oFileCopyMap{$strPathKey}{$strFileKey}{size} = $lFileSize;
$oFileCopyMap{$strPathKey}{$strFileKey}{modification_time} =
$oBackupManifest->numericGet($strSectionFile, $strFile, MANIFEST_SUBKEY_TIMESTAMP, false);
$oFileCopyMap{$strPathKey}{$strFileKey}{checksum} =
2015-01-21 17:44:08 +02:00
$oBackupManifest->get($strSectionFile, $strFile, MANIFEST_SUBKEY_CHECKSUM, false);
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
}
# pg_control should always be in the backup (unless this is an offline backup)
if (!defined($oFileCopyMap{&MANIFEST_KEY_BASE}{&FILE_PG_CONTROL}) && optionGet(OPTION_ONLINE))
{
confess &log(ERROR, "global/pg_control must be present in all online backups\n" .
'HINT: Is something wrong with the clock or filesystem timestamps?', ERROR_FILE_MISSING);
}
# If there are no files to backup then we'll exit with a warning unless in test mode. The other way this could happen is if
# the database is down and backup is called with --no-online twice in a row.
if ($lFileTotal == 0)
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
if (!optionGet(OPTION_TEST))
{
confess &log(ERROR, "no files have changed since the last backup - this seems unlikely", ERROR_FILE_MISSING);
}
}
else
{
# Create backup and result queues
my $oResultQueue = Thread::Queue->new();
my @oyBackupQueue;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Variables used for local copy
my $lSizeCurrent = 0; # Running total of bytes copied
my $bCopied; # Was the file copied?
my $lCopySize; # Size reported by copy
my $strCopyChecksum; # Checksum reported by copy
# Determine how often the manifest will be saved
my $lManifestSaveCurrent = 0;
my $lManifestSaveSize = int($lSizeTotal / 100);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
if (optionSource(OPTION_MANIFEST_SAVE_THRESHOLD) ne SOURCE_DEFAULT ||
$lManifestSaveSize < optionGet(OPTION_MANIFEST_SAVE_THRESHOLD))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
$lManifestSaveSize = optionGet(OPTION_MANIFEST_SAVE_THRESHOLD);
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Start backup test point
&log(TEST, TEST_BACKUP_START);
# Iterate all backup files
foreach my $strPathKey (sort(keys(%oFileCopyMap)))
{
if (optionGet(OPTION_THREAD_MAX) > 1)
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
$oyBackupQueue[@oyBackupQueue] = Thread::Queue->new();
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
foreach my $strFileKey (sort {$b cmp $a} (keys(%{$oFileCopyMap{$strPathKey}})))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
my $oFileCopy = $oFileCopyMap{$strPathKey}{$strFileKey};
# Skip files marked to be copied later
next if $$oFileCopy{skip};
if (optionGet(OPTION_THREAD_MAX) > 1)
{
$oyBackupQueue[@oyBackupQueue - 1]->enqueue($oFileCopy);
}
else
{
# Backup the file
($bCopied, $lSizeCurrent, $lCopySize, $strCopyChecksum) =
backupFile($self->{oFile}, $$oFileCopy{db_file}, $$oFileCopy{backup_file}, $bCompress,
$$oFileCopy{checksum}, $$oFileCopy{modification_time},
$$oFileCopy{size}, $lSizeTotal, $lSizeCurrent);
$lManifestSaveCurrent = backupManifestUpdate($oBackupManifest, $$oFileCopy{file_section}, $$oFileCopy{file},
$bCopied, $lCopySize, $strCopyChecksum, $lManifestSaveSize,
$lManifestSaveCurrent);
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
}
# If multi-threaded then create threads to copy files
if (optionGet(OPTION_THREAD_MAX) > 1)
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
# Load module dynamically
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
require pgBackRest::Protocol::ThreadGroup;
pgBackRest::Protocol::ThreadGroup->import();
for (my $iThreadIdx = 0; $iThreadIdx < optionGet(OPTION_THREAD_MAX); $iThreadIdx++)
{
my %oParam;
$oParam{compress} = $bCompress;
$oParam{size_total} = $lSizeTotal;
$oParam{queue} = \@oyBackupQueue;
$oParam{result_queue} = $oResultQueue;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Keep the protocol layer from timing out
protocolGet()->keepAlive();
threadGroupRun($iThreadIdx, 'backup', \%oParam);
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Keep the protocol layer from timing out
protocolGet()->keepAlive();
# Start backup test point
&log(TEST, TEST_BACKUP_START);
# Complete thread queues
my $bDone = false;
do
{
$bDone = threadGroupComplete();
# Read the messages that are passed back from the backup threads
while (my $oMessage = $oResultQueue->dequeue_nb())
{
&log(TRACE, "message received in master queue: section = $$oMessage{file_section}, file = $$oMessage{file}" .
", copied = $$oMessage{copied}");
2014-05-13 18:23:15 +03:00
$lManifestSaveCurrent = backupManifestUpdate($oBackupManifest, $$oMessage{file_section}, $$oMessage{file},
$$oMessage{copied}, $$oMessage{size}, $$oMessage{checksum},
$lManifestSaveSize, $lManifestSaveCurrent);
}
# Keep the protocol layer from timing out
protocolGet()->keepAlive();
}
while (!$bDone);
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
# Copy pg_control last - this is required for backups taken during recovery
my $oFileCopy = $oFileCopyMap{&MANIFEST_KEY_BASE}{&FILE_PG_CONTROL};
if (defined($oFileCopy))
{
my ($bCopied, $lSizeCurrent, $lCopySize, $strCopyChecksum) =
backupFile($self->{oFile}, $$oFileCopy{db_file}, $$oFileCopy{backup_file}, $bCompress,
$$oFileCopy{checksum}, $$oFileCopy{modification_time},
$$oFileCopy{size});
backupManifestUpdate($oBackupManifest, $$oFileCopy{file_section}, $$oFileCopy{file},
$bCopied, $lCopySize, $strCopyChecksum);
$lSizeTotal += $$oFileCopy{size};
}
# Return from function and log return values if any
return logDebugReturn
(
$strOperation,
{name => 'lSizeTotal', value => $lSizeTotal}
);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
####################################################################################################################################
# process
2014-06-04 18:58:30 +03:00
#
# Process the database backup.
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
####################################################################################################################################
sub process
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
my $self = shift;
# Assign function parameters, defaults, and log debug info
my
(
$strOperation
) =
logDebugParam
(
OP_BACKUP_PROCESS
);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Record timestamp start
my $lTimestampStart = time();
# Store local type, compress, and hardlink options since they can be modified by the process
my $strType = optionGet(OPTION_TYPE);
my $bCompress = optionGet(OPTION_COMPRESS);
my $bHardLink = optionGet(OPTION_HARDLINK);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Not supporting remote backup hosts yet
if ($self->{oFile}->isRemote(PATH_BACKUP))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
confess &log(ERROR, 'remote backup host not currently supported');
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
# Create the cluster backup path
$self->{oFile}->pathCreate(PATH_BACKUP_CLUSTER, undef, undef, true);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Load or build backup.info
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
my $oBackupInfo = new pgBackRest::BackupInfo($self->{oFile}->pathGet(PATH_BACKUP_CLUSTER));
2015-01-20 21:42:22 +02:00
# Build backup tmp and config
my $strBackupTmpPath = $self->{oFile}->pathGet(PATH_BACKUP_TMP);
my $strBackupConfFile = $self->{oFile}->pathGet(PATH_BACKUP_TMP, 'backup.manifest');
2015-01-20 21:42:22 +02:00
# Declare the backup manifest
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
my $oBackupManifest = new pgBackRest::Manifest($strBackupConfFile, false);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Find the previous backup based on the type
2015-02-03 01:33:11 +02:00
my $oLastManifest = undef;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
my $strBackupLastPath = $self->typeFind($strType, $self->{oFile}->pathGet(PATH_BACKUP_CLUSTER));
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
if (defined($strBackupLastPath))
{
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
$oLastManifest = new pgBackRest::Manifest($self->{oFile}->pathGet(PATH_BACKUP_CLUSTER) . "/${strBackupLastPath}/backup.manifest");
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
&log(INFO, 'last backup label = ' . $oLastManifest->get(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_LABEL) .
', version = ' . $oLastManifest->get(INI_SECTION_BACKREST, INI_KEY_VERSION));
# If this is incr or diff warn if certain options have changed
if ($strType ne BACKUP_TYPE_FULL)
{
my $strKey;
# Warn if compress option changed
if (!$oLastManifest->boolTest(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_COMPRESS, undef, $bCompress))
{
&log(WARN, "${strType} backup cannot alter compress option to '" . boolFormat($bCompress) .
"', reset to value in ${strBackupLastPath}");
$bCompress = $oLastManifest->boolGet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_COMPRESS);
}
# Warn if hardlink option changed
if (!$oLastManifest->boolTest(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_HARDLINK, undef, $bHardLink))
{
&log(WARN, "${strType} backup cannot alter hardlink option to '" . boolFormat($bHardLink) .
"', reset to value in ${strBackupLastPath}");
$bHardLink = $oLastManifest->boolGet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_HARDLINK);
}
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
else
{
if ($strType eq BACKUP_TYPE_DIFF || $strType eq BACKUP_TYPE_INCR)
{
&log(WARN, "no prior backup exists, ${strType} backup has been changed to full");
}
$strType = BACKUP_TYPE_FULL;
}
2015-01-21 01:00:03 +02:00
# Backup settings
$oBackupManifest->set(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_TYPE, undef, $strType);
$oBackupManifest->numericSet(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_TIMESTAMP_START, undef, $lTimestampStart);
$oBackupManifest->boolSet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_COMPRESS, undef, $bCompress);
$oBackupManifest->boolSet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_HARDLINK, undef, $bHardLink);
$oBackupManifest->boolSet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_ONLINE, undef, optionGet(OPTION_ONLINE));
$oBackupManifest->boolSet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_ARCHIVE_COPY, undef,
!optionGet(OPTION_ONLINE) ||
(optionGet(OPTION_BACKUP_ARCHIVE_CHECK) && optionGet(OPTION_BACKUP_ARCHIVE_COPY)));
$oBackupManifest->boolSet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_ARCHIVE_CHECK, undef,
!optionGet(OPTION_ONLINE) || optionGet(OPTION_BACKUP_ARCHIVE_CHECK));
# Database info
my ($fDbVersion, $iControlVersion, $iCatalogVersion, $ullDbSysId) =
$self->{oDb}->info($self->{oFile}, optionGet(OPTION_DB_PATH));
$oBackupInfo->check($fDbVersion, $iControlVersion, $iCatalogVersion, $ullDbSysId);
$oBackupManifest->set(MANIFEST_SECTION_BACKUP_DB, MANIFEST_KEY_DB_VERSION, undef, $fDbVersion);
$oBackupManifest->numericSet(MANIFEST_SECTION_BACKUP_DB, MANIFEST_KEY_CONTROL, undef, $iControlVersion);
$oBackupManifest->numericSet(MANIFEST_SECTION_BACKUP_DB, MANIFEST_KEY_CATALOG, undef, $iCatalogVersion);
$oBackupManifest->numericSet(MANIFEST_SECTION_BACKUP_DB, MANIFEST_KEY_SYSTEM_ID, undef, $ullDbSysId);
# Start backup (unless --no-online is set)
my $strArchiveStart;
my $oTablespaceMap;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Don't start the backup but do check if PostgreSQL is running
if (!optionGet(OPTION_ONLINE))
{
if ($self->{oFile}->exists(PATH_DB_ABSOLUTE, optionGet(OPTION_DB_PATH) . '/' . FILE_POSTMASTER_PID))
{
if (optionGet(OPTION_FORCE))
{
&log(WARN, '--no-online passed and ' . FILE_POSTMASTER_PID . ' exists but --force was passed so backup will ' .
2014-12-19 00:05:06 +02:00
'continue though it looks like the postmaster is running and the backup will probably not be ' .
'consistent');
}
else
{
confess &log(ERROR, '--no-online passed but ' . FILE_POSTMASTER_PID . ' exists - looks like the postmaster is ' .
'running. Shutdown the postmaster and try again, or use --force.', ERROR_POSTMASTER_RUNNING);
}
}
}
# Else start the backup normally
else
{
2015-01-21 01:00:03 +02:00
my $strTimestampDbStart;
# Start the backup
2015-01-21 01:00:03 +02:00
($strArchiveStart, $strTimestampDbStart) =
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
$self->{oDb}->backupStart($self->{oFile}, optionGet(OPTION_DB_PATH), BACKREST_NAME . ' Backup Started at ' .
timestampFormat(undef, $lTimestampStart), optionGet(OPTION_START_FAST));
2015-01-21 01:00:03 +02:00
# Record the archive start location
2015-01-21 01:00:03 +02:00
$oBackupManifest->set(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_ARCHIVE_START, undef, $strArchiveStart);
&log(INFO, "archive start: ${strArchiveStart}");
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Build the backup manifest
$oTablespaceMap = $self->{oDb}->tablespaceMapGet();
}
# Buid the manifest
$oBackupManifest->build($self->{oFile}, optionGet(OPTION_DB_PATH), $oLastManifest, optionGet(OPTION_ONLINE),
$oTablespaceMap);
2014-07-28 01:13:23 +03:00
&log(TEST, TEST_MANIFEST_BUILD);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Check if an aborted backup exists for this stanza
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
if (-e $strBackupTmpPath)
{
my $bUsable = false;
my $strReason = "resume is disabled";
my $oAbortedManifest;
# Attempt to read the manifest file in the aborted backup to seeif it can be used. If any error at all occurs then the
# backup will be considered unusable and a resume will not be attempted.
if (optionGet(OPTION_RESUME))
{
$strReason = "unable to read ${strBackupTmpPath}/backup.manifest";
eval
{
# Load the aborted manifest
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
$oAbortedManifest = new pgBackRest::Manifest("${strBackupTmpPath}/backup.manifest");
# Key and values that do not match
my $strKey;
my $strValueNew;
my $strValueAborted;
# Check version
if ($oBackupManifest->get(INI_SECTION_BACKREST, INI_KEY_VERSION) ne
$oAbortedManifest->get(INI_SECTION_BACKREST, INI_KEY_VERSION))
2014-10-14 22:44:50 +03:00
{
$strKey = INI_KEY_VERSION;
$strValueNew = $oBackupManifest->get(INI_SECTION_BACKREST, INI_KEY_VERSION);
$strValueAborted = $oAbortedManifest->get(INI_SECTION_BACKREST, INI_KEY_VERSION);
}
# Check format
elsif ($oBackupManifest->get(INI_SECTION_BACKREST, INI_KEY_FORMAT) ne
$oAbortedManifest->get(INI_SECTION_BACKREST, INI_KEY_FORMAT))
{
$strKey = INI_KEY_FORMAT;
$strValueNew = $oBackupManifest->get(INI_SECTION_BACKREST, INI_KEY_FORMAT);
$strValueAborted = $oAbortedManifest->get(INI_SECTION_BACKREST, INI_KEY_FORMAT);
}
# Check backup type
elsif ($oBackupManifest->get(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_TYPE) ne
$oAbortedManifest->get(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_TYPE))
{
$strKey = MANIFEST_KEY_TYPE;
$strValueNew = $oBackupManifest->get(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_TYPE);
$strValueAborted = $oAbortedManifest->get(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_TYPE);
}
# Check prior label
elsif ($oBackupManifest->get(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_PRIOR, undef, false, '<undef>') ne
$oAbortedManifest->get(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_PRIOR, undef, false, '<undef>'))
{
$strKey = MANIFEST_KEY_PRIOR;
$strValueNew = $oBackupManifest->get(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_PRIOR, undef, false, '<undef>');
$strValueAborted = $oAbortedManifest->get(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_PRIOR, undef, false, '<undef>');
}
# Check compression
elsif ($oBackupManifest->boolGet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_COMPRESS) ne
$oAbortedManifest->boolGet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_COMPRESS))
{
$strKey = MANIFEST_KEY_COMPRESS;
$strValueNew = $oBackupManifest->boolGet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_COMPRESS);
$strValueAborted = $oAbortedManifest->boolGet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_COMPRESS);
2014-10-14 22:44:50 +03:00
}
# Check hardlink
elsif ($oBackupManifest->boolGet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_HARDLINK) ne
$oAbortedManifest->boolGet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_HARDLINK))
{
$strKey = MANIFEST_KEY_HARDLINK;
$strValueNew = $oBackupManifest->boolGet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_HARDLINK);
$strValueAborted = $oAbortedManifest->boolGet(MANIFEST_SECTION_BACKUP_OPTION, MANIFEST_KEY_HARDLINK);
}
# If key is defined then something didn't match
if (defined($strKey))
{
$strReason = "new ${strKey} '${strValueNew}' does not match aborted ${strKey} '${strValueAborted}'";
}
# Else the backup can be resumed
else
{
$bUsable = true;
}
};
}
2014-06-04 18:58:30 +03:00
# If the aborted backup is usable then clean it
if ($bUsable)
{
&log(WARN, 'aborted backup of same type exists, will be cleaned to remove invalid files and resumed');
&log(TEST, TEST_BACKUP_RESUME);
# Clean the old backup tmp path
$self->tmpClean($oBackupManifest, $oAbortedManifest);
}
# Else remove it
else
{
&log(WARN, "aborted backup exists, but cannot be resumed (${strReason}) - will be dropped and recreated");
&log(TEST, TEST_BACKUP_NORESUME);
remove_tree($self->{oFile}->pathGet(PATH_BACKUP_TMP))
or confess &log(ERROR, "unable to delete tmp path: ${strBackupTmpPath}");
$self->{oFile}->pathCreate(PATH_BACKUP_TMP);
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
# Else create the backup tmp path
else
{
logDebugMisc($strOperation, "create temp backup path ${strBackupTmpPath}");
$self->{oFile}->pathCreate(PATH_BACKUP_TMP);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
2015-01-20 21:42:22 +02:00
# Save the backup manifest
$oBackupManifest->save();
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Perform the backup
my $lBackupSizeTotal = $self->processManifest($strType, $bCompress, $bHardLink, $oBackupManifest);
&log(INFO, "${strType} backup size = " . fileSizeFormat($lBackupSizeTotal));
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Stop backup (unless --no-online is set)
my $strArchiveStop;
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
if (optionGet(OPTION_ONLINE))
{
2015-01-21 01:00:03 +02:00
my $strTimestampDbStop;
($strArchiveStop, $strTimestampDbStop) = $self->{oDb}->backupStop();
2015-01-21 01:00:03 +02:00
$oBackupManifest->set(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_ARCHIVE_STOP, undef, $strArchiveStop);
&log(INFO, 'archive stop: ' . $strArchiveStop);
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# If archive logs are required to complete the backup, then check them. This is the default, but can be overridden if the
# archive logs are going to a different server. Be careful of this option because there is no way to verify that the backup
# will be consistent - at least not here.
if (optionGet(OPTION_ONLINE) && optionGet(OPTION_BACKUP_ARCHIVE_CHECK))
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
{
2015-01-20 21:42:22 +02:00
# Save the backup manifest a second time - before getting archive logs in case that fails
$oBackupManifest->save();
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Create the modification time for the archive logs
my $lModificationTime = time();
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# After the backup has been stopped, need to make a copy of the archive logs need to make the db consistent
logDebugMisc($strOperation, "retrieve archive logs ${strArchiveStart}:${strArchiveStop}");
New simpler configuration and consistent project/exe/path naming. * The repo-path option now always refers to the repository where backups and archive are stored, whether local or remote, so the repo-remote-path option has been removed. The new spool-path option can be used to define a location for queueing WAL segments when archiving asynchronously. Otherwise, a local repository is no longer required. * Implemented a new config format which should be far simpler to use. See the User Guide and Configuration Reference for details but for a simple configuration all options can now be placed in the stanza section. Options that are shared between stanzas can be placed in the [global] section. More complex configurations can still make use of command sections though this should be a rare use case. * The default configuration filename is now pgbackrest.conf instead of pg_backrest.conf. This was done for consistency with other naming changes but also to prevent old config files from being loaded accidentally. * The default repository name was changed from /var/lib/backup to /var/lib/pgbackrest. * Lock files are now stored in /tmp/pgbackrest by default. These days /run/pgbackrest would be the preferred location but that would require init scripts which are not part of this release. The lock-path option can be used to configure the lock directory. * Log files are now stored in /var/log/pgbackrest by default and no longer have the date appended so they can be managed with logrotate. The log-path option can be used to configure the lock directory. * Executable filename changed from pg_backrest to pgbackrest.
2016-04-14 15:30:54 +02:00
my $oArchive = new pgBackRest::Archive();
my $strArchiveId = $oArchive->getCheck($self->{oFile});
my @stryArchive = $oArchive->range($strArchiveStart, $strArchiveStop, $fDbVersion < 9.3);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
foreach my $strArchive (@stryArchive)
{
my $strArchiveFile = $oArchive->walFileName($self->{oFile}, $strArchiveId, $strArchive, false, 600);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
if (optionGet(OPTION_BACKUP_ARCHIVE_COPY))
{
logDebugMisc($strOperation, "archive: ${strArchive} (${strArchiveFile})");
# Copy the log file from the archive repo to the backup
my $strDestinationFile = "base/pg_xlog/${strArchive}" . ($bCompress ? ".$self->{oFile}->{strCompressExtension}" : '');
my $bArchiveCompressed = $strArchiveFile =~ "^.*\.$self->{oFile}->{strCompressExtension}\$";
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
my ($bCopyResult, $strCopyChecksum, $lCopySize) =
$self->{oFile}->copy(PATH_BACKUP_ARCHIVE, "${strArchiveId}/${strArchiveFile}",
PATH_BACKUP_TMP, $strDestinationFile,
$bArchiveCompressed, $bCompress,
undef, $lModificationTime, undef, true);
# Add the archive file to the manifest so it can be part of the restore and checked in validation
my $strPathSection = 'base:path';
my $strPathLog = 'pg_xlog';
my $strFileSection = 'base:file';
my $strFileLog = "pg_xlog/${strArchive}";
# Compare the checksum against the one already in the archive log name
if ($strArchiveFile !~ "^${strArchive}-${strCopyChecksum}(\\.$self->{oFile}->{strCompressExtension}){0,1}\$")
{
confess &log(ERROR, "error copying WAL segment '${strArchiveFile}' to backup - checksum recorded with " .
"file does not match actual checksum of '${strCopyChecksum}'", ERROR_CHECKSUM);
}
# Set manifest values
$oBackupManifest->set($strFileSection, $strFileLog, MANIFEST_SUBKEY_USER,
$oBackupManifest->get($strPathSection, $strPathLog, MANIFEST_SUBKEY_USER));
$oBackupManifest->set($strFileSection, $strFileLog, MANIFEST_SUBKEY_GROUP,
$oBackupManifest->get($strPathSection, $strPathLog, MANIFEST_SUBKEY_GROUP));
$oBackupManifest->set($strFileSection, $strFileLog, MANIFEST_SUBKEY_MODE, '0700');
$oBackupManifest->set($strFileSection, $strFileLog, MANIFEST_SUBKEY_TIMESTAMP, $lModificationTime);
$oBackupManifest->set($strFileSection, $strFileLog, MANIFEST_SUBKEY_SIZE, $lCopySize);
$oBackupManifest->set($strFileSection, $strFileLog, MANIFEST_SUBKEY_CHECKSUM, $strCopyChecksum);
}
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
}
# Create the path for the new backup
my $lTimestampStop = time();
my $strBackupLabel = backupLabelFormat($strType, $strBackupLastPath, $lTimestampStop);
# Record timestamp stop in the config
$oBackupManifest->set(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_TIMESTAMP_STOP, undef, $lTimestampStop + 0);
$oBackupManifest->set(MANIFEST_SECTION_BACKUP, MANIFEST_KEY_LABEL, undef, $strBackupLabel);
2015-01-20 21:42:22 +02:00
# Save the backup manifest final time
$oBackupManifest->save();
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
&log(INFO, "new backup label = ${strBackupLabel}");
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Rename the backup tmp path to complete the backup
logDebugMisc($strOperation, "move ${strBackupTmpPath} to " . $self->{oFile}->pathGet(PATH_BACKUP_CLUSTER, $strBackupLabel));
$self->{oFile}->move(PATH_BACKUP_TMP, undef, PATH_BACKUP_CLUSTER, $strBackupLabel);
# Create a link to the most recent backup
$self->{oFile}->remove(PATH_BACKUP_CLUSTER, "latest");
$self->{oFile}->linkCreate(PATH_BACKUP_CLUSTER, $strBackupLabel, PATH_BACKUP_CLUSTER, "latest", undef, true);
# Save backup info
$oBackupInfo->add($self->{oFile}, $oBackupManifest);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
# Return from function and log return values if any
return logDebugReturn
(
$strOperation
);
v0.10: Backup and archiving are functional This version has been put into production at Resonate, so it does work, but there are a number of major caveats. * No restore functionality, but the backup directories are consistent Postgres data directories. You'll need to either uncompress the files or turn off compression in the backup. Uncompressed backups on a ZFS (or similar) filesystem are a good option because backups can be restored locally via a snapshot to create logical backups or do spot data recovery. * Archiving is single-threaded. This has not posed an issue on our multi-terabyte databases with heavy write volume. Recommend a large WAL volume or to use the async option with a large volume nearby. * Backups are multi-threaded, but the Net::OpenSSH library does not appear to be 100% threadsafe so it will very occasionally lock up on a thread. There is an overall process timeout that resolves this issue by killing the process. Yes, very ugly. * Checksums are lost on any resumed backup. Only the final backup will record checksum on multiple resumes. Checksums from previous backups are correctly recorded and a full backup will reset everything. * The backup.manifest is being written as Storable because Config::IniFile does not seem to handle large files well. Would definitely like to save these as human-readable text. * Absolutely no documentation (outside the code). Well, excepting these release notes. * Lots of other little things and not so little things. Much refactoring to follow.
2014-03-06 03:53:13 +03:00
}
2014-06-04 18:58:30 +03:00
1;