Barman (backup and recovery manager) is an administration tool for disaster recovery of PostgreSQL servers written in Python. Barman can perform remote backups of multiple servers in business critical environments, and helps DBAs during the recovery phase.

Barman’s most wanted features include: backup catalogues, incremental backup, retention policies, remote recovery, archiving and compression of WAL files and of backups. Barman is written and maintained by PostgreSQL professionals 2ndQuadrant.

Introduction

In a perfect world, there would be no need for a backup. However it is important, especially in business environments, to be prepared for when the “unexpected” happens. In a database scenario, the unexpected could take any of the following forms:

  • data corruption;
  • system failure, including hardware failure;
  • human error;
  • natural disaster.

In such cases, any ICT manager or DBA should be able to repair the incident and recover the database in the shortest possible time. We normally refer to this discipline as Disaster recovery.

This guide assumes that you are familiar with theoretical disaster recovery concepts, and you have a grasp of PostgreSQL fundamentals in terms of physical backup and disaster recovery. If not, we encourage you to read the PostgreSQL documentation or any of the recommended books on PostgreSQL.

Professional training on this topic is another effective way of learning these concepts. At any time of the year you can find many courses available all over the world, delivered by PostgreSQL companies such as 2ndQuadrant.

For now, you should be aware that any PostgreSQL physical/binary backup (not to be confused with the logical backups produced by the pg_dump utility) is composed of:

  • a base backup;
  • one or more WAL files (usually collected through continuous archiving).

PostgreSQL offers the core primitives that allow DBAs to setup a really robust Disaster Recovery environment. However, it becomes complicated to manage multiple backups, from one or more PostgreSQL servers. Restoring a given backup is another task that any PostgreSQL DBA would love to see more automated and user friendly.

With these goals in mind, 2ndQuadrant started the development of Barman for PostgreSQL. Barman is an acronym for “Backup and Recovery Manager”. Currently Barman works only on Linux and Unix operating systems.

Before you start

The first step is to decide the architecture of your backup. In a simple scenario, you have one PostgreSQL instance (server) running on a host. You want your data continuously backed up to another server, called the backup server.

Barman allows you to launch PostgreSQL backups directly from the backup server, using SSH connections. Furthermore, it allows you to centralise your backups in case you have more than one PostgreSQL server to manage.

During this guide, we will assume that:

  • there is one PostgreSQL instance on a host (called pg for simplicity)
  • there is one backup server on another host (called backup)
  • communication via SSH between the two servers is enabled
  • the PostgreSQL server can be reached from the backup server as the postgres operating system user (or another user with PostgreSQL database superuser privileges, typically configured via ident authentication)

It is important to note that, for disaster recovery, these two servers must not share any physical resource except for the network. You can use Barman in geographical redundancy scenarios for better disaster recovery outcomes.

System requirements

  • Linux/Unix
  • Python 2.6 or 2.7
  • Python modules:
    • argcomplete
    • argh >= 0.21.2
    • psycopg2
    • python-dateutil < 2.0 (since version 2.0 requires python3)
    • distribute (optional)
  • PostgreSQL >= 8.3
  • rsync >= 3.0.4

Important: PostgreSQL’s Point-In-Time-Recovery requires the same major version of the sourcePostgreSQL server to be installed on the backup server.

Important: Users of RedHat Enterprise Linux, CentOS and Scientific Linux are required to install the Extra Packages Enterprise Linux (EPEL) repository.

Note: Version 1.2.3 of Barman has been refactored for Python 3 support. Please consider it as experimental now and report any bug through the ticketing system on SourceForge or mailing list.

Installation

Important: The recommended way to install Barman is by using the available packages for your GNU/Linux distribution.

On RedHat/CentOS using RPM packages

Barman can be installed on RHEL7, RHEL6 and RHEL5 Linux systems using RPM packages. It is required to install the Extra Packages Enterprise Linux (EPEL) repository beforehand.

RPM packages for Barman are available via Yum through the PostgreSQL Global Development Group RPM repository. You need to follow the instructions for your distribution (RedHat, CentOS, Fedora, etc.) and architecture as detailed at yum.postgresql.org.

Then, as root simply type:

yum install barman

2ndQuadrant also maintains RPM packages for Barman and distributes them through Sourceforge.net.

On Debian/Ubuntu using packages

Barman can be installed on Debian and Ubuntu Linux systems using packages.

It is directly available in the official repository for Debian Sid (unstable) and Ubuntu 14.04 (Trusty Tahr).

However, the recommended method for installing Barman on Debian and Ubuntu is through the PostgreSQL Community APT repository. Instructions can be found in the APT section of the PostgreSQL Wiki.

Note: Thanks to the direct involvement of Barman developers in the PostgreSQL Community APT repository project, you will have access to the most updated versions of Barman.

Installing Barman is as simple as typing, as root user:

apt-get install barman

From sources

WARNING: Manual installation of Barman from sources should only be performed by expert GNU/Linux users. Installing Barman this way requires system administration activities such as dependencies management, barman user creation, configuration of the barman.conf file, cron setup for the barman cron command, log management, etc.

Create a system user called barman on the backup server. As barman user, download the sources and uncompress them.

For a system-wide installation, type:

barman@backup$ ./setup.py build
# run this command with root privileges or through sudo
barman@backup# ./setup.py install

For a local installation, type:

barman@backup$ ./setup.py install --user

Important: The --user option works only with python-distribute

barman will be installed in your user directory (make sure that your PATH environment variable is set properly).

Getting started

Prerequisites

SSH connection

Barman needs a bidirectional SSH connection between the barman user on the backup server and the postgres user. SSH must be configured such that there is no password prompt presented when connecting.

As the barman user on the backup server, generate an SSH key with an empty password, and append the public key to the authorized_keys file of the postgres user on the pg server.

The barman user on the backup server should then be able to perform the following operation without typing a password:

barman@backup$ ssh postgres@pg

The procedure must be repeated with sides swapped in order to allow the postgres user on the pg server to connect to the backup server as the barman user without typing a password:

postgres@pg$ ssh barman@backup

For further information, refer to OpenSSH documentation.

PostgreSQL connection

You need to make sure that the backup server allows connection to the PostgreSQL server on pg as superuser (postgres).

You can choose your favourite client authentication method among those offered by PostgreSQL. More information can be found in the PostgreSQL Documentation.

barman@backup$ psql -c 'SELECT version()' -U postgres -h pg

Note: As of version 1.1.2, Barman honours the application_name connection option for PostgreSQL servers 9.0 or higher.

Backup directory

Barman needs a main backup directory to store all the backups. Even though you can define a separate folder for each server you want to back up and for each type of resource (backup or WAL segments, for instance), we suggest that you adhere to the default rules and stick with the conventions that Barman chooses for you.

You will see that the configuration file (as explained below) defines a barman_home variable, which is the directory where Barman will store all your backups by default. We choose /var/lib/barman as home directory for Barman:

barman@backup$ sudo mkdir /var/lib/barman
barman@backup$ sudo chown barman:barman /var/lib/barman

Important: We assume that you have enough space, and that you have already thought about redundancy and safety of your disks.

Basic configuration

In the docs directory you will find a minimal configuration file. Use it as a template, and copy it to /etc/barman.conf, or to ~/.barman.conf. In general, the former applies to all the users on the backup server, while the latter applies only to the barman user; for the purpose of this tutorial there is no difference in using one or the other.

From version 1.2.1, you can use /etc/barman/barman.conf as default system configuration file.

The configuration file follows the standard INI format, and is split in:

  • a section for general configuration (identified by the barman label)
  • a section for each PostgreSQL server to be backed up (identified by the server label, e.g. main or pg)1

Global/server options

Every option in the configuration file has a scope:

  • global
  • server
  • global/server

Global options can be present in the general section (identified by barman). Server options can only be specified in a server section.

Some options can be defined at global level and overridden at server level, allowing users to specify a generic behaviour and refine it for one or more servers. For a list of all the available configurations and their scope, please refer to section 5 of the man page.

man 5 barman

Configuration files directory

As of version 1.1.2, you can now specify a directory for configuration files similarly to other Linux applications, using the configuration_files_directory option (empty by default). If the value of configuration_files_directory is a directory, Barman will read all the files with .conf extension that exist in that folder. For example, if you set it to /etc/barman.d, you can specify your PostgreSQL servers placing each section in a separate .conf file inside the /etc/barman.d folder.

Otherwise, you can use Barman’s standard way of specifying sections within the main configuration file.

Lock files

Since version 1.5.0, Barman allows DBAs to specify a directory for lock files through the barman_lock_directory global option.

Lock files are used to coordinate concurrent work at global and server level (for example, cron operations, backup operations, access to the WAL archive, etc.).

By default (for backward compatibility reasons), barman_lock_directory is set to barman_home.

Important: This change won’t affect users upgrading from a version of Barman older than 1.5.0, unless you have written applications that depend on the names of the lock files. However, this is not a typical and common case for Barman and most of users do not fall into this category.

Tip: Users are encouraged to use a directory in a volatile partition, such as the one dedicated to run-time variable data (e.g. /var/run/barman).

Example of configuration

Here follows a basic example of PostgreSQL configuration:

[barman]
barman_home = /var/lib/barman
barman_user = barman
log_file = /var/log/barman/barman.log
compression = gzip
reuse_backup = link
minimum_redundancy = 1

[main]
description = "Main PostgreSQL Database"
ssh_command = ssh postgres@pg
conninfo = host=pg user=postgres

For more detailed information, please refer to the distributed barman.conf file.

Initial checks

Once you have created your configuration file (or files), you can now test Barman’s configuration by executing:

barman@backup$ barman show-server main
barman@backup$ barman check main

Write down the incoming_wals_directory, as printed by the barman show-server main command, because you will need it to setup continuous WAL archiving.

Important: The barman check main command automatically creates all the directories for the continuous backup of the main server.

Continuous WAL archiving

Edit the postgresql.conf file of the PostgreSQL instance on the pg database and activate the archive mode:

wal_level = 'archive' # For PostgreSQL >= 9.0
archive_mode = on
archive_command = 'rsync -a %p barman@backup:INCOMING_WALS_DIRECTORY/%f'

Make sure you change the INCOMING_WALS_DIRECTORY placeholder with the value returned by the barman show-server main command above.

In case you use Hot Standby, wal_level must be set to hot_standby.

Restart the PostgreSQL server.

In order to test that continuous archiving is on and properly working, you need to check both the PostgreSQL server and the backup server (in particular, that WAL files are correctly collected in the destination directory).

Warning: It currently is a requirement that WAL files from PostgreSQL are shipped to the Barman server. Without archive_command being properly set in PostgreSQL to send WAL files to Barman, full backups cannot be taken.

Listing the servers

The following command displays the list of all the available servers:

barman@backup$ barman list-server

Executing a full backup

To take a backup for the main server, issue the following command:

barman@backup$ barman backup main

As of version 1.1.0, you can serialise the backup of your managed servers by using the all target for the server:

barman@backup$ barman backup all

This will iterate through your available servers and sequentially take a backup for each of them.

Immediate Checkpoint

As of version 1.3.0, it is possible to use the immediate_checkpoint configuration global/server option (set to false by default).

Before starting a backup, Barman requests a checkpoint, which generates additional workload. Normally that checkpoint is throttled according to the settings for workload control on the PostgreSQL server, which means that the backup could be delayed.

If immediate_checkpoint is set to true, PostgreSQL will not try to limit the workload, and the checkpoint will happen at maximum speed, starting the backup as soon as possible.

At any time, you can override the configuration option behaviour, by issuing barman backup with any of these two options:

  • --immediate-checkpoint, which forces an immediate checkpoint;
  • --no-immediate-checkpoint, which forces to wait for the checkpoint to happen.

Viewing the list of backups for a server

To list all the available backups for a given server, issue:

barman@backup$ barman list-backup main

the format of the output is as in:

main - 20120529T092136 - Wed May 30 15:20:25 2012 - Size: 5.0 TiB
 - WAL Size: 845.0 GiB (tablespaces: a:/disk1/a, t:/disk2/t)

where 20120529T092136 is the ID of the backup and Wed May 30 15:20:25 2012 is the start time of the operation, Size is the size of the base backup and WAL Size is the size of the archived WAL files.

As of version 1.1.2, you can get a listing of the available backups for all your servers, using the all target for the server:

barman@backup$ barman list-backup all

Restoring a whole server

To restore a whole server issue the following command:

barman@backup$ barman recover main 20110920T185953 /path/to/recover/dir

where 20110920T185953 is the ID of the backup to be restored. When this command completes successfully, /path/to/recover/dir contains a complete data directory ready to be started as a PostgreSQL database server.

Here is an example of a command that starts the server:

barman@backup$ pg_ctl -D /path/to/recover/dir start

Important: If you run this command as user barman, it will become the database superuser.

You can retrieve a list of backup IDs for a specific server with:

barman@backup$ barman list-backup srvpgsql

Important: Barman does not currently keep track of symbolic links inside PGDATA (except for tablespaces inside pg_tblspc). We encourage system administrators to keep track of symbolic links and to add them to the disaster recovery plans/procedures in case they need to be restored in their original location.

Remote recovery

Barman is able to recover a backup on a remote server through the --remote-ssh-command COMMAND option for the recover command.

If this option is specified, barman uses COMMAND to connect to a remote host.

Note: The postgres user is normally used to recover on a remote host.

There are some limitations when using remote recovery. It is important to be aware that:

  • unless get-wal is specified in the recovery_options (available from version 1.5.0), Barman requires at least 4GB of free space in the system’s temporary directory (usually /tmp);
  • the SSH connection between Barman and the remote host must use public key exchange authentication method;
  • the remote user must be able to create the required destination directories for PGDATA and, where applicable, tablespaces;
  • there must be enough free space on the remote server to contain the base backup and the WAL files needed for recovery.

Relocating one or more tablespaces

Important: As of version 1.3.0, it is possible to relocate a tablespace both with local and remote recovery.

Barman is able to automatically relocate one or more tablespaces using the recover command with the --tablespace option. The option accepts a pair of values as arguments using the NAME:DIRECTORY format:

  • name/identifier of the tablespace (NAME);
  • destination directory (DIRECTORY).

If the destination directory does not exists, Barman will try to create it (assuming you have enough privileges).

Restoring to a given point in time

Barman employs PostgreSQL’s Point-in-Time Recovery (PITR) by allowing DBAs to specify a recovery target, either as a timestamp or as a transaction ID; you can also specify whether the recovery target should be included or not in the recovery.

The recovery target can be specified using one of three mutually exclusive options:

  • --target-time TARGET_TIME: to specify a timestamp
  • --target-xid TARGET_XID: to specify a transaction ID
  • --target-name TARGET_NAME: to specify a named restore point - previously created with the pg_create_restore_point(name) function2

You can use the --exclusive option to specify whether to stop immediately before or immediately after the recovery target.

Barman allows you to specify a target timeline for recovery, using the target-tli option. The notion of timeline goes beyond the scope of this document; you can find more details in the PostgreSQL documentation, or in one of 2ndQuadrant’s Recovery training courses.

Retry of copy in backup/recovery operations

As of version 1.3.3, it is possible to take advantage of two new options in Barman:

  • basebackup_retry_times (set to 0 by default)
  • basebackup_retry_sleep (set to 30 by default)

When issuing a backup or a recovery, Barman normally tries to copy the base backup once. If the copy fails (e.g. due to network problems), Barman terminates the operation with a failure.

By setting basebackup_retry_times, Barman will try to re-execute a copy operation as many times as requested by the user. The basebackup_retry_sleep option specifies the number of seconds that Barman will wait between each attempt.

At any time you can override the configuration option behaviour from the command line, when issuing barman backup or barman recover, using:

  • --retry-times <retry_number> (same logic as basebackup_retry_times)
  • --no-retry (same as --retry-times 0)
  • --retry-sleep <number_of_seconds> (same logic as basebackup_retry_sleep)

Available commands

Barman commands are applied to three different levels:

  • general commands, which apply to the backup catalogue
  • server commands, which apply to a specific server (list available backups, execute a backup, etc.)
  • backup commands, which apply to a specific backup in the catalogue (display information, issue a recovery, delete the backup, etc.)

In the following sections the available commands will be described in detail.

General commands

List available servers

You can display the list of active servers that have been configured for your backup system with:

barman list-server

Maintenance mode

You can perform maintenance operations, like compressing WAL files and moving them from the incoming directory to the archived one, with:

barman cron

This command enforces retention policies on those servers that have:

  • retention_policy not empty and valid;
  • retention_policy_mode set to auto.

Note: This command should be executed in a cron script. Our recommendation is to schedule barman cron to run every minute.

Server commands

Show the configuration for a given server

You can show the configuration parameters for a given server with:

barman show-server <server_name>

Take a base backup

You can perform a full backup (base backup) for a given server with:

barman backup [--immediate-checkpoint] <server_name>

Tip: You can use barman backup all to sequentially backup all your configured servers.

Show available backups for a server

You can list the catalogue of available backups for a given server with:

barman list-backup <server_name>

Check a server is properly working

You can check if the connection to a given server is properly working with:

barman check <server_name>

Tip: You can use barman check all to check all your configured servers.

From version 1.3.3, you can automatically be notified if the latest backup of a given server is older than, for example, 7 days.3

Barman introduces the option named last_backup_maximum_age having the following syntax:

last_backup_maximum_age = {value {DAYS | WEEKS | MONTHS}}

where value is a positive integer representing the number of days, weeks or months of the time frame.

Diagnose a Barman installation

You can gather important information about all the configured server using:

barman diagnose

The diagnose command also provides other useful information, such as global configuration, SSH version, Python version, rsync version, as well as current configuration and status of all servers.

Tip: You can use barman diagnose when you want to ask questions or report errors to Barman developers, providing them with all the information about your issue.

Rebuild the WAL archive

At any time, you can regenerate the content of the WAL archive for a specific server (or every server, using the all shortcut). The WAL archive is contained in the xlog.db file, and every Barman server has its own copy. From version 1.2.4 you can now rebuild the xlog.db file with the rebuild-xlogdb command. This will scan all the archived WAL files and regenerate the metadata for the archive.

Important: Users of Barman < 1.2.3 might have suffered from a bug due to bad locking in highly concurrent environments. You can now regenerate the WAL archive using the rebuild-xlogdb command.

barman rebuild-xlogdb <server_name>

Access WAL archive using get-wal

From version 1.5.0, Barman allows users to request any xlog file from its WAL archive through the get-wal command:

barman get-wal [-o OUTPUT_DIRECTORY] [-j|-x] <server_name> <wal_id>

If the requested WAL file is found in the server archive, the uncompressed content will be returned to STDOUT, unless otherwise specified.

The following options are available for the get-wal command:

  • -o allows users to specify a destination directory where Barman will deposit the requested WAL file
  • -j will compress the output using bzip2 algorithm
  • -x will compress the output using gzip algorithm

It is possible to use get-wal during a recovery operation, transforming the Barman server in a WAL hub for your servers. This can be automatically achieved by adding the get-wal value to the recovery_options global/server configuration option:

recovery_options = 'get-wal'

recovery_options is a global/server option that accepts a list of comma separated values. If the keyword get-wal is present, during a recovery operation Barman will prepare the recovery.conf file by setting the restore_command so that barman get-wal is used to fetch the required WAL files.

This is an example of a restore_command for a remote recovery:

restore_command = 'ssh barman@pgbackup barman get-wal SERVER %f > %p'

This is an example of a restore_command for a local recovery:

restore_command = 'barman get-wal SERVER %f > %p'

Important: Even though recovery_options aims to automate the process, using the get-wal facility requires manual intervention and proper testing.

Backup commands

Note: Remember: a backup ID can be retrieved with barman list-backup <server_name>

Show backup information

You can show all the available information for a particular backup of a given server with:

barman show-backup <server_name> <backup_id>

From version 1.1.2, in order to show the latest backup, you can issue:

barman show-backup <server_name> latest

Delete a backup

You can delete a given backup with:

barman delete <server_name> <backup_id>

From version 1.1.2, in order to delete the oldest backup, you can issue:

barman delete <server_name> oldest

List backup files

You can list the files (base backup and required WAL files) for a given backup with:

barman list-files [--target TARGET_TYPE] <server_name> <backup_id>

With the --target TARGET_TYPE option, it is possible to choose the content of the list for a given backup.

Possible values for TARGET_TYPE are:

  • data: lists just the data files;
  • standalone: lists the base backup files, including required WAL files;
  • wal: lists all WAL files from the beginning of the base backup to the start of the following one (or until the end of the log);
  • full: same as data + wal.

The default value for TARGET_TYPE is standalone.

Important: The list-files command facilitates interaction with external tools, and therefore can be extremely useful to integrate > Barman into your archiving procedures.

Main features

Incremental backup

From version 1.4.0, Barman implements file-level incremental backup. Incremental backup is a kind of full periodic backup which saves only data changes from the latest full backup available in the catalogue for a specific PostgreSQL server. It must not be confused with differential backup, which is implemented by WAL continuous archiving.

The main goals of incremental backup in Barman are:

  • Reduce the time taken for the full backup process
  • Reduce the disk space occupied by several periodic backups (data deduplication)

This feature heavily relies on rsync and hard links, which must be therefore supported by both the underlying operating system and the file system where the backup data resides.

The main concept is that a subsequent base backup will share those files that have not changed since the previous backup, leading to relevant savings in disk usage. This is particularly true of VLDB contexts and, more in general, of those databases containing a high percentage of read-only historical tables.

Barman implements incremental backup through a global/server option, called reuse_backup, that transparently manages the barman backup command. It accepts three values:

  • off: standard full backup (default)
  • link: incremental backup, by reusing the last backup for a server and creating a hard link of the unchanged files (for backup space and time reduction)
  • copy: incremental backup, by reusing the last backup for a server and creating a copy of the unchanged files (just for backup time reduction)

The most common scenario is to set reuse_backup to link, as follows:

reuse_backup = link

Setting this at global level will automatically enable incremental backup for all your servers.

As a final note, users can override the setting of the reuse_backup option through the --reuse-backup runtime option for the barman backup command. Similarly, the runtime option accepts three values: off, link and copy. For example, you can run a one-off incremental backup as follows:

barman backup --reuse-backup=link <server_name>

WAL compression

The barman cron command (see below) will compress WAL files if the compression option is set in the configuration file. This option allows three values:

  • gzip: for Gzip compression (requires gzip)
  • bzip2: for Bzip2 compression (requires bzip2)
  • custom: for custom compression, which requires you to set the following options as well:
    • custom_compression_filter: a compression filter
    • custom_decompression_filter: a decompression filter

Limiting bandwidth usage

From version 1.2.1, it is possible to limit the usage of I/O bandwidth through the bandwidth_limit option (global/per server), by specifying the maximum number of kilobytes per second. By default it is set to 0, meaning no limit.

In case you have several tablespaces and you prefer to limit the I/O workload of your backup procedures on one or more tablespaces, you can use the tablespace_bandwidth_limit option (global/per server):

tablespace_bandwidth_limit = tbname:bwlimit[, tbname:bwlimit, ...]

The option accepts a comma separated list of pairs made up of the tablespace name and the bandwidth limit (in kilobytes per second).

When backing up a server, Barman will try and locate any existing tablespace in the above option. If found, the specified bandwidth limit will be enforced. If not, the default bandwidth limit for that server will be applied.

Network Compression

From version 1.3.0 it is possible to reduce the size of transferred data using compression. It can be enabled using the network_compression option (global/per server):

network_compression = true|false

Setting this option to true will enable data compression during network transfers (for both backup and recovery). By default it is set to false.

Backup ID shortcuts

As of version 1.1.2, you can use any of the following shortcuts to identify a particular backup for a given server:

  • latest: the latest available backup for that server, in chronological order. You can also use the last synonym.
  • oldest: the oldest available backup for that server, in chronological order. You can also use the first synonym.

These aliases can be used with any of the following commands: show-backup, delete, list-files and recover.

Minimum redundancy safety

From version 1.2.0, you can define the minimum number of periodic backups for a PostgreSQL server.

You can use the global/per server configuration option called minimum_redundancy for this purpose, by default set to 0.

By setting this value to any number greater than 0, Barman makes sure that at any time you will have at least that number of backups in a server catalogue.

This will protect you from accidental barman delete operations.

Important: Make sure that your policy retention settings do not collide with minimum redundancy requirements. Regularly check Barman’s log for messages on this topic.

Retention policies

From version 1.2.0, Barman supports retention policies for backups.

A backup retention policy is an user-defined policy that determines how long backups and related archive logs (Write Ahead Log segments) need to be retained for recovery procedures.

Based on the user’s request, Barman retains the periodic backups required to satisfy the current retention policy, and any archived WAL files required for the complete recovery of those backups.

Barman users can define a retention policy in terms of backup redundancy (how many periodic backups) or a recovery window (how long).

Retention policy based on redundancy

In a redundancy based retention policy, the user determines how many periodic backups to keep. A redundancy-based retention policy is contrasted with retention policies that use a recovery window.

Retention policy based on recovery window

A recovery window is one type of Barman backup retention policy, in which the DBA specifies a period of time and Barman ensures retention of backups and/or archived WAL files required for point-in-time recovery to any time during the recovery window. The interval always ends with the current time and extends back in time for the number of days specified by the user. For example, if the retention policy is set for a recovery window of seven days, and the current time is 9:30 AM on Friday, Barman retains the backups required to allow point-in-time recovery back to 9:30 AM on the previous Friday.

Scope

Retention policies can be defined for:

  • PostgreSQL periodic base backups: through the retention_policy configuration option;
  • Archive logs, for Point-In-Time-Recovery: through the wal_retention_policy configuration option.

Important: In a temporal dimension, archive logs must be included in the time window of periodic backups.

There are two typical use cases here: full or partial point-in-time recovery.

Full point in time recovery scenario

Base backups and archive logs share the same retention policy, allowing DBAs to recover at any point in time from the first available backup.

Partial point in time recovery scenario

Base backup retention policy is wider than that of archive logs, allowing users for example to keep full weekly backups of the last 6 months, but archive logs for the last 4 weeks (granting to recover at any point in time starting from the last 4 periodic weekly backups).

Important: Currently, Barman implements only the full point in time recovery scenario, by constraining the wal_retention_policy option to main.

How they work

Retention policies in Barman can be:

  • automated: enforced by barman cron;
  • manual: Barman simply reports obsolete backups and allows DBAs to delete them.

Important: Currently Barman does not implement manual enforcement. This feature will be available in future versions.

Configuration and syntax

Retention policies can be defined through the following configuration options:

  • retention_policy: for base backup retention;
  • wal_retention_policy: for archive logs retention;
  • retention_policy_mode: can only be set to auto (retention policies are automatically enforced by the barman cron command).

These configuration options can be defined both at a global level and a server level, allowing users maximum flexibility on a multi-server environment.

Syntax for retention_policy

The general syntax for a base backup retention policy through retention_policy is the following:

retention_policy = {REDUNDANCY value | RECOVERY WINDOW OF value {DAYS | WEEKS | MONTHS}}

Where:

  • syntax is case insensitive;
  • value is an integer and is > 0;
  • in case of redundancy retention policy:
    • value must be greater than or equal to the server minimum redundancy level (if not is is assigned to that value and a warning is generated);
    • the first valid backup is the value-th backup in a reverse ordered time series;
  • in case of recovery window policy:
    • the point of recoverability is: current time - window;
    • the first valid backup is the first available backup before the point of recoverability; its value in a reverse ordered time series must be greater than or equal to the server minimum redundancy level (if not is is assigned to that value and a warning is generated).

By default, retention_policy is empty (no retention enforced).

Syntax for wal_retention_policy

Currently, the only allowed value for wal_retention_policy is the special value main, that maps the retention policy of archive logs to that of base backups.

Concurrent Backup and backup from a standby

Normally, during backup operations, Barman uses PostgreSQL native functions pg_start_backup and pg_stop_backup for exclusive backup. These operations are not allowed on a read-only standby server.

As of version 1.3.1, Barman is also capable of performing backups of PostgreSQL 9.2/9.3 database servers in a concurrent way, primarily through the backup_options configuration parameter.4

This introduces a new architecture scenario with Barman: backup from a standby server, using rsync.

Important: Concurrent backup requires users of PostgreSQL 9.2 and 9.3 to install the pgespresso open source extension on the PostgreSQL server. For more detailed information and the source code, please visit the pgespresso extension website.

By default, backup_options is transparently set to exclusive_backup (the only supported method by any Barman version prior to 1.3.1).

When backup_options is set to concurrent_backup, Barman activates the concurrent backup mode for a server and follows these two simple rules:

  • ssh_command must point to the destination Postgres server;
  • conninfo must point to a database on the destination Postgres 9.2 or 9.3 server where pgespresso is correctly installed through CREATE EXTENSION.

The destination Postgres server can be either the master or a streaming replicated standby server.

Note: When backing up from a standby server, continuous archiving of WAL files must be configured on the master to ship files to the Barman server (as outlined in the “Continuous WAL archiving” section above)5.

Hook scripts

Barman allows a database administrator to run hook scripts on these two events:

  • before and after a backup
  • before and after a WAL file is archived

There are two types of hook scripts that Barman can manage:

  • standard hook scripts (already present in Barman since version 1.1.0)
  • retry hook scripts, introduced in version 1.5.0

The only difference between these two types of hook scripts is that Barman executes a standard hook script only once, without checking its return code, whereas a retry hook script may be executed more than once depending on its return code.

Precisely, when executing a retry hook script, Barman checks the return code and retries indefinitely until the script returns either SUCCESS (with standard return code 0), or ABORT_CONTINUE (return code 62), or ABORT_STOP (return code 63). Barman treats any other return code as a transient failure to be retried. Users are given more power: a hook script can control its workflow by specifying whether a failure is transient. Also, in case of a ‘pre’ hook script, by returning ABORT_STOP, users can request Barman to interrupt the main operation with a failure.

Hook scripts are executed in the following order:

  1. The standard ‘pre’ hook script (if present)
  2. The retry ‘pre’ hook script (if present)
  3. The actual event (i.e. backup operation, or WAL archiving), if retry ‘pre’ hook script was not aborted with ABORT_STOP
  4. The retry ‘post’ hook script (if present)
  5. The standard ‘post’ hook script (if present)

The output generated by any hook script is written in the log file of Barman.

Note: Currently, ABORT_STOP is ignored by retry ‘post’ hook scripts. In these cases, apart from lodging an additional warning, ABORT_STOP will behave like ABORT_CONTINUE.

Backup scripts

Version 1.1.0 introduced backup scripts.

These scripts can be configured with the following global configuration options (which can be overridden on a per server basis):

  • pre_backup_script: hook script executed before a base backup, only once, with no check on the exit code
  • pre_backup_retry_script: retry hook script executed before a base backup, repeatedly until success or abort
  • post_backup_retry_script: retry hook script executed after a base backup, repeatedly until success or abort
  • post_backup_script: hook script executed after a base backup, only once, with no check on the exit code

The script definition is passed to a shell and can return any exit code. Only in case of a retry script, Barman checks the return code (see the upper section).

The shell environment will contain the following variables:

  • BARMAN_BACKUP_DIR: backup destination directory
  • BARMAN_BACKUP_ID: ID of the backup
  • BARMAN_CONFIGURATION: configuration file used by barman
  • BARMAN_ERROR: error message, if any (only for the post phase)
  • BARMAN_PHASE: phase of the script, either pre or post
  • BARMAN_PREVIOUS_ID: ID of the previous backup (if present)
  • BARMAN_RETRY: 1 if it is a retry script (from 1.5.0), 0 if not
  • BARMAN_SERVER: name of the server
  • BARMAN_STATUS: status of the backup
  • BARMAN_VERSION: version of Barman (from 1.2.1)

WAL archive scripts

Version 1.3.0 introduced WAL archive hook scripts.

Similarly to backup scripts, archive scripts can be configured with global configuration options (which can be overridden on a per server basis):

  • pre_archive_script: hook script executed before a WAL file is archived by maintenance (usually barman cron), only once, with no check on the exit code
  • pre_archive_retry_script: retry hook script executed before a WAL file is archived by maintenance (usually barman cron), repeatedly until success or abort
  • post_archive_retry_script: retry hook script executed after a WAL file is archived by maintenance, repeatedly until success or abort
  • post_archive_script: hook script executed after a WAL file is archived by maintenance, only once, with no check on the exit code

The script is executed through a shell and can return any exit code. Only in case of a retry script, Barman checks the return code (see the upper section).

Archive scripts share with backup scripts some environmental variables:

  • BARMAN_CONFIGURATION: configuration file used by barman
  • BARMAN_ERROR: error message, if any (only for the post phase)
  • BARMAN_PHASE: phase of the script, either pre or post
  • BARMAN_SERVER: name of the server

Following variables are specific to archive scripts:

  • BARMAN_SEGMENT: name of the WAL file
  • BARMAN_FILE: full path of the WAL file
  • BARMAN_SIZE: size of the WAL file
  • BARMAN_TIMESTAMP: WAL file timestamp
  • BARMAN_COMPRESSION: type of compression used for the WAL file

Integration with standby servers

Barman has been designed for integration with standby servers (with streaming replication or traditional file based log shipping) and high availability tools like repmgr.

From an architectural point of view, PostgreSQL must be configured to archive WAL files directly to the Barman server.

Support and sponsor opportunities

Barman is free software, written and maintained by 2ndQuadrant. If you require support on using Barman, or if you need new features, please get in touch with 2ndQuadrant. You can sponsor the development of new features of Barman and PostgreSQL which will be made publicly available as open source.

For further information, please visit:

Important: When submitting requests on the mailing list, please always report the output of the barman diagnose command.

Submitting a bug

Barman has been extensively tested, and is currently being used in several production environments. However, as any software, Barman is not bug free.

If you discover a bug, please follow this procedure:

  • execute the barman diagnose command;
  • file a bug through the Sourceforge bug tracker, by attaching the output obtained by the diagnostics command above (barman diagnose).

Authors

In alphabetical order:

  • Gabriele Bartolini (project leader)
  • Stefano Bianucci (developer)
  • Giuseppe Broccolo (QA/testing)
  • Giulio Calacoci (developer)
  • Francesco Canovai (QA/testing)
  • Gianni Ciolli (QA/testing)
  • Marco Nenciarini (lead developer)

Past contributors:

  • Carlo Ascani

Links

  • check-barman: a Nagios plugin for Barman, written by Holger Hamann (MIT license)
  • puppet-barman: Barman module for Puppet (GPL)

License and Contributions

Barman is the exclusive property of 2ndQuadrant Italia and its code is distributed under GNU General Public License 3.

Copyright (C) 2011-2015 2ndQuadrant.it.

Barman has been partially funded through 4CaaSt, a research project funded by the European Commission’s Seventh Framework programme.

Contributions to Barman are welcome, and will be listed in the AUTHORS file. 2ndQuadrant Italia requires that any contributions provide a copyright assignment and a disclaimer of any work-for-hire ownership claims from the employer of the developer. This lets us make sure that all of the Barman distribution remains free code. Please contact info@2ndQuadrant.it for a copy of the relevant Copyright Assignment Form.


  1. all and barman are reserved words and cannot be used as server labels.

  2. Only available on PostgreSQL 9.1 and above.

  3. This feature is commonly known among the development team members as smelly backup check.

  4. Concurrent backup is a technology that has been available in PostgreSQL since version 9.1, through the streaming replication protocol (using, for example, a tool like pg_basebackup).

  5. In case of concurrent backup, currently Barman does not have a way to determine that the closing WAL file of a full backup has actually been shipped - opposite to the case of an exclusive backup where it is Postgres itself that makes sure that the WAL file is correctly archived. Be aware that the full backup cannot be considered consistent until that WAL file has been received and archived by Barman. We encourage Barman users to wait to delete the previous backup - at least until that moment.