repmgr is a suite of open-source tools to manage replication and failover within a cluster of PostgreSQL servers. It enhances PostgreSQL's built-in replication capabilities with utilities to set up standby servers, monitor replication, and perform administrative tasks such as failover or switchover operations.

The repmgr suite provides two main tools:

• repmgr - a command-line tool used to perform administrative tasks such as:

  • setting up standby servers
  • promoting a standby server to master
  • switching over master and standby servers
  • displaying the status of servers in the replication cluster

• repmgrd is a daemon which actively monitors servers in a replication cluster and performs the following tasks:

  • monitoring and recording replication performance
  • performing failover by detecting failure of the master and promoting the most suitable standby server
  • provide notifications about events in the cluster to a user-defined script which can perform tasks such as sending alerts by email

repmgr supports and enhances PostgreSQL's built-in streaming replication, which provides a single read/write master server and one or more read-only standbys containing near-real time copies of the master server's database.

For a multi-master replication solution, please see 2ndQuadrant's BDR (bi-directional replication) extension.

http://2ndquadrant.com/en-us/resources/bdr/

For selective replication, e.g. of individual tables or databases from one server to another, please see 2ndQuadrant's pglogical extension.

http://2ndquadrant.com/en-us/resources/pglogical/

This guide assumes that you are familiar with PostgreSQL administration and streaming replication concepts. For further details on streaming replication, see this link:

http://www.postgresql.org/docs/current/interactive/warm-standby.html#STREAMING-REPLICATION

The following terms are used throughout the repmgr documentation.

• replication cluster

In the repmgr documentation, "replication cluster" refers to the network of PostgreSQL servers connected by streaming replication.

• node

A node is a server within a replication cluster.

• upstream node

This is the node a standby server is connected to; either the master server or in the case of cascading replication, another standby.

• failover

This is the action which occurs if a master server fails and a suitable standby is promoted as the new master. The repmgrd daemon supports automatic failover to minimise downtime.

• switchover

In certain circumstances, such as hardware or operating system maintenance, it's necessary to take a master server offline; in this case a controlled switchover is necessary, whereby a suitable standby is promoted and the existing master removed from the replication cluster in a controlled manner. The repmgr command line client provides this functionality.

• witness server

repmgr provides functionality to set up a so-called "witness server" to assist in determining a new master server in a failover situation with more than one standby. The witness server itself is not part of the replication cluster, although it does contain a copy of the repmgr metadata schema (see below).

The purpose of a witness server is to provide a "casting vote" where servers in the replication cluster are split over more than one location. In the event of a loss of connectivity between locations, the presence or absence of the witness server will decide whether a server at that location is promoted to master; this is to prevent a "split-brain" situation where an isolated location interprets a network outage as a failure of the (remote) master and promotes a (local) standby.

A witness server only needs to be created if repmgrd is in use.

In order to effectively manage a replication cluster, repmgr needs to store information about the servers in the cluster in a dedicated database schema. This schema is automatically created during the first step in initialising a repmgr-controlled cluster (repmgr master register) and contains the following objects:

tables:

• repl_events: records events of interest
• repl_nodes: connection and status information for each server in the replication cluster
• repl_monitor: historical standby monitoring information written by repmgrd

views:

• repl_show_nodes: based on the table repl_nodes, additionally showing the name of the server's upstream node
• repl_status: when repmgrd's monitoring is enabled, shows current monitoring status for each node

The repmgr metadata schema can be stored in an existing database or in its own dedicated database.

A dedicated database superuser is required to own the meta-database as well as carry out administrative actions.

repmgr is developed and tested on Linux and OS X, but should work on any UNIX-like system supported by PostgreSQL itself.

Current versions of repmgr support PostgreSQL from version 9.3. If you are interested in using repmgr on earlier versions of PostgreSQL you can download version 2.1 which supports PostgreSQL from version 9.1.

All servers in the replication cluster must be running the same major version of PostgreSQL, and we recommend that they also run the same minor version.

The repmgr tools must be installed on each server in the replication cluster.

A dedicated system user for repmgr is not required; as many repmgr and repmgrd actions require direct access to the PostgreSQL data directory, it should be executed by the postgres user.

Additionally, we recommend installing rsync and enabling passwordless ssh connectivity between all servers in the replication cluster.

We recommend installing repmgr using the available packages for your system.

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

• Debian/Ubuntu: the most recent repmgr packages are available from the PostgreSQL Community APT repository ( http://apt.postgresql.org/ ). Instructions can be found in the APT section of the PostgreSQL Wiki ( https://wiki.postgresql.org/wiki/Apt ).

See PACKAGES.md for details on building .deb and .rpm packages from the repmgr source code.

repmgr source code can be obtained directly from the project GitHub repository:

git clone https://github.com/2ndQuadrant/repmgr

Release tarballs are also available:

https://github.com/2ndQuadrant/repmgr/releases
http://repmgr.org/downloads.php

repmgr is compiled in the same way as a PostgreSQL extension using the PGXS infrastructure, e.g.:

sudo make USE_PGXS=1 install

repmgr can be built from source in any environment suitable for building PostgreSQL itself.

repmgr and repmgrd use a common configuration file, by default called repmgr.conf (although any name can be used if explicitly specified). At the very least, repmgr.conf must contain the connection parameters for the local repmgr database; see repmgr configuration file below for more details.

The configuration file will be searched for in the following locations:

• a configuration file specified by the -f/--config-file command line option
• repmgr.conf in the local directory
• /etc/repmgr.conf
• the directory reported by pg_config --sysconfdir

Note that if a file is explicitly specified with -f/--config-file, an error will be raised if it is not found or not readable and no attempt will be made to check default locations; this is to prevent repmgr reading the wrong file.

For a full list of annotated configuration items, see the file repmgr.conf.sample.

The following parameters in the configuration file can be overridden with command line options:

• -L/--log-level
• -b/--pg_bindir

The following section will describe how to set up a basic replication cluster with a master and a standby server using the repmgr command line tool. It is assumed PostgreSQL is installed on both servers in the cluster, rsync is available and password-less SSH connections are possible between both servers.

TIP: for testing repmgr, it's possible to use multiple PostgreSQL instances running on different ports on the same computer, with password-less SSH access to localhost enabled.

On the master server, a PostgreSQL instance must be initialised and running. The following replication settings must be included in postgresql.conf:

# Ensure WAL files contain enough information to enable read-only queries
# on the standby

wal_level = 'hot_standby'

# Enable up to 10 replication connections

max_wal_senders = 10

# How much WAL to retain on the master to allow a temporarily
# disconnected standby to catch up again. The larger this is, the
# longer the standby can be disconnected. This is needed only in
# 9.3; from 9.4, replication slots can be used instead (see below).

wal_keep_segments = 5000

# Enable read-only queries on a standby
# (Note: this will be ignored on a master but we recommend including
# it anyway)

hot_standby = on

# If archive_mode is enabled, check that 'archive_command' is non empty
# (however it's not practical to check that it actually represents a valid
# command).
# 
# From PostgreSQL 9.5, archive_mode can be one of 'off', 'on' or 'always'
# so for ease of backwards compatibility, rather than explicitly check for an
# enabled mode, check that it's not "off".
archive_mode = on

# Set archive command to a script or application that will safetly store
# you WALs in a secure place. /bin/true is an example of a command that
# ignores archiving. Use something more sensible.
archive_command = '/bin/true'

TIP: rather than editing these settings in the default postgresql.conf file, create a separate file such as postgresql.replication.conf and include it from the end of the main configuration file with: include 'postgresql.replication.conf'

Create a dedicated PostgreSQL superuser account and a database for the repmgr metadata, e.g.

createuser -s repmgr
createdb repmgr -O repmgr

For the examples in this document, the name repmgr will be used for both user and database, but any names can be used.

Ensure the repmgr user has appropriate permissions in pg_hba.conf and can connect in replication mode; pg_hba.conf should contain entries similar to the following:

local   replication   repmgr                              trust
host    replication   repmgr      127.0.0.1/32            trust
host    replication   repmgr      192.168.1.0/24          trust

local   repmgr        repmgr                              trust
host    repmgr        repmgr      127.0.0.1/32            trust
host    repmgr        repmgr      192.168.1.0/24          trust

Adjust according to your network environment and authentication requirements.

On the standby, do not create a PostgreSQL instance, but do ensure an empty directory is available for the postgres system user to create a data directory.

Create a repmgr.conf file on the master server. The file must contain at least the following parameters:

cluster=test
node=1
node_name=node1
conninfo='host=repmgr_node1 user=repmgr dbname=repmgr'

• cluster: an arbitrary name for the replication cluster; this must be identical on all nodes
• node: a unique integer identifying the node
• node_name: a unique string identifying the node; we recommend a name specific to the server (e.g. 'server_1'); avoid names indicating the current replication role like 'master' or 'standby' as the server's role could change.
• conninfo: a valid connection string for the repmgr database on the current server. (On the standby, the database will not yet exist, but repmgr needs to know the connection details to complete the setup process).

repmgr.conf should not be stored inside the PostgreSQL data directory, as it could be overwritten when setting up or reinitialising the PostgreSQL server. See section Configuration above for further details about repmgr.conf.

repmgr will create a schema named after the cluster and prefixed with repmgr_, e.g. repmgr_test; we also recommend that you set the repmgr user's search path to include this schema name, e.g.

ALTER USER repmgr SET search_path TO repmgr_test, "$user", public;

To enable repmgr to support a replication cluster, the master node must be registered with repmgr, which creates the repmgr database and adds a metadata record for the server:

$ repmgr -f repmgr.conf master register
[2016-01-07 16:56:46] [NOTICE] master node correctly registered for cluster test with id 1 (conninfo: host=repmgr_node1 user=repmgr dbname=repmgr)

The metadata record looks like this:

repmgr=# SELECT * FROM repmgr_test.repl_nodes;
 id |  type   | upstream_node_id | cluster | name  |                  conninfo                   | slot_name | priority | active
----+---------+------------------+---------+-------+---------------------------------------------+-----------+----------+--------
  1 | master  |                  | test    | node1 | host=repmgr_node1 dbname=repmgr user=repmgr |           |      100 | t
(1 row)

Each server in the replication cluster will have its own record and will be updated when its status or role changes.

Create a repmgr.conf file on the standby server. It must contain at least the same parameters as the master's repmgr.conf, but with the values node, node_name and conninfo adjusted accordingly, e.g.:

cluster=test
node=2
node_name=node2
conninfo='host=repmgr_node2 user=repmgr dbname=repmgr'

Clone the standby with:

$ repmgr -h repmgr_node1 -U repmgr -d repmgr -D /path/to/node2/data/ -f /etc/repmgr.conf standby clone
[2016-01-07 17:21:26] [NOTICE] destination directory '/path/to/node2/data/' provided
[2016-01-07 17:21:26] [NOTICE] starting backup...
[2016-01-07 17:21:26] [HINT] this may take some time; consider using the -c/--fast-checkpoint option
NOTICE:  pg_stop_backup complete, all required WAL segments have been archived
[2016-01-07 17:21:28] [NOTICE] standby clone (using pg_basebackup) complete
[2016-01-07 17:21:28] [NOTICE] you can now start your PostgreSQL server
[2016-01-07 17:21:28] [HINT] for example : pg_ctl -D /path/to/node2/data/ start

This will clone the PostgreSQL data directory files from the master at repmgr_node1 using PostgreSQL's pg_basebackup utility. A recovery.conf file containing the correct parameters to start streaming from this master server will be created automatically, and unless otherwise the postgresql.conf and pg_hba.conf files will be copied from the master.

Make any adjustments to the PostgreSQL configuration files now, then start the standby server.

NOTE: repmgr standby clone does not require repmgr.conf, however we recommend providing this as repmgr will set the application_name parameter in recovery.conf as the value provided in node_name, making it easier to identify the node in pg_stat_replication. It's also possible to provide some advanced options for controlling the standby cloning process; see next section for details.

Connect to the master server and execute:

repmgr=# SELECT * FROM pg_stat_replication;
-[ RECORD 1 ]----+------------------------------
pid              | 7704
usesysid         | 16384
usename          | repmgr
application_name | node2
client_addr      | 192.168.1.2
client_hostname  |
client_port      | 46196
backend_start    | 2016-01-07 17:32:58.322373+09
backend_xmin     |
state            | streaming
sent_location    | 0/3000220
write_location   | 0/3000220
flush_location   | 0/3000220
replay_location  | 0/3000220
sync_priority    | 0
sync_state       | async

Register the standby server with:

repmgr -f /etc/repmgr.conf standby register
[2016-01-08 11:13:16] [NOTICE] standby node correctly registered for cluster test with id 2 (conninfo: host=repmgr_node2 user=repmgr dbname=repmgr)

Connect to the standby server's repmgr database and check the repl_nodes table:

repmgr=# SELECT * FROM repmgr_test.repl_nodes ORDER BY id;
 id |  type   | upstream_node_id | cluster | name  |                  conninfo                   | slot_name | priority | active
----+---------+------------------+---------+-------+---------------------------------------------+-----------+----------+--------
  1 | master  |                  | test    | node1 | host=repmgr_node1 dbname=repmgr user=repmgr |           |      100 | t
  2 | standby |                1 | test    | node2 | host=repmgr_node2 dbname=repmgr user=repmgr |           |      100 | t
(2 rows)

The standby server now has a copy of the records for all servers in the replication cluster. Note that the relationship between master and standby is explicitly defined via the upstream_node_id value, which shows here that the standby's upstream server is the replication cluster master. While of limited use in a simple master/standby replication cluster, this information is required to effectively manage cascading replication (see below).

The above section demonstrates the simplest possible way to cloneb a standby server. Depending on your circumstances, finer-grained controlover the cloning process may be necessary.

By default, pg_basebackup performs a checkpoint before beginning the backup process. However, a normal checkpoint may take some time to complete; a fast checkpoint can be forced with the -c/--fast-checkpoint option. However this may impact performance of the server being cloned from so should be used with care.

Further options can be passed to the pg_basebackup utility via the setting pg_basebackup_options in repmgr.conf. See the PostgreSQL documentation for more details of available options: http://www.postgresql.org/docs/current/static/app-pgbasebackup.html

By default repmgr uses the pg_basebackup utility to clone a standby's data directory from the master. Under some circumstances it may be desirable to use rsync to do this, such as when resyncing the data directory of a failed server with an active replication node.

To use rsync instead of pg_basebackup, provide the -r/--rsync-only option when executing repmgr standby clone.

Note that repmgr forces rsync to use --checksum mode to ensure that all the required files are copied. This results in additional I/O on both source and destination server as the contents of files existing on both servers need to be compared, meaning this method is not necessarily faster than making a fresh clone with pg_basebackup.

By default, repmgr will attempt to copy the standard configuration files (postgresql.conf, pg_hba.conf and pg_ident.conf) even if they are located outside of the data directory (though currently they will be copied into the standby's data directory). To prevent this happening, when executing repmgr standby clone provide the --ignore-external-config-files option.

If using rsync to clone a standby, additional control over which files not to transfer is possible by configuring rsync_options in repmgr.conf, which enables any valid rsync options to be passed to that command, e.g.:

rsync_options='--exclude=postgresql.local.conf'

Cascading replication, introduced with PostgreSQL 9.2, enables a standby server to replicate from another standby server rather than directly from the master, meaning replication changes "cascade" down through a hierarchy of servers. This can be used to reduce load on the master and minimize bandwith usage between sites.

repmgr supports cascading replication. When cloning a standby, in repmgr.conf set the parameter upstream_node to the id of the server the standby should connect to, and repmgr will perform the clone using this server and create recovery.conf to point to it. Note that if upstream_node is not explicitly provided, repmgr will use the master as the server to clone from.

To demonstrate cascading replication, ensure you have a master and standby set up as shown above in the section "Setting up a simple replication cluster with repmgr". Create an additional standby server with repmgr.conf looking like this:

cluster=test
node=3
node_name=node3
conninfo='host=repmgr_node3 user=repmgr dbname=repmgr'
upstream_node=2

Ensure upstream_node contains the node id of the previously created standby. Clone this standby (using the connection parameters for the existing standby) and register it:

$ repmgr -h repmgr_node2 -U repmgr -d repmgr -D /path/to/node3/data/ -f /etc/repmgr.conf standby clone
[2016-01-08 13:44:52] [NOTICE] destination directory 'node_3/data/' provided
[2016-01-08 13:44:52] [NOTICE] starting backup (using pg_basebackup)...
[2016-01-08 13:44:52] [HINT] this may take some time; consider using the -c/--fast-checkpoint option
[2016-01-08 13:44:52] [NOTICE] standby clone (using pg_basebackup) complete
[2016-01-08 13:44:52] [NOTICE] you can now start your PostgreSQL server
[2016-01-08 13:44:52] [HINT] for example : pg_ctl -D /path/to/node_3/data start

$ repmgr -f /etc/repmgr.conf standby register
[2016-01-08 14:04:32] [NOTICE] standby node correctly registered for cluster test with id 3 (conninfo: host=repmgr_node3 dbname=repmgr user=repmgr)

After starting the standby, the repl_nodes table will look like this:

repmgr=# SELECT * FROM repmgr_test.repl_nodes ORDER BY id;
 id |  type   | upstream_node_id | cluster | name  |                  conninfo                   | slot_name | priority | active
----+---------+------------------+---------+-------+---------------------------------------------+-----------+----------+--------
  1 | master  |                  | test    | node1 | host=repmgr_node1 dbname=repmgr user=repmgr |           |      100 | t
  2 | standby |                1 | test    | node2 | host=repmgr_node2 dbname=repmgr user=repmgr |           |      100 | t
  3 | standby |                2 | test    | node3 | host=repmgr_node3 dbname=repmgr user=repmgr |           |      100 | t
(3 rows)

Replication slots were introduced with PostgreSQL 9.4 and are designed to ensure that any standby connected to the master using a replication slot will always be able to retrieve the required WAL files. This removes the need to manually manage WAL file retention by estimating the number of WAL files that need to be maintained on the master using wal_keep_segments. Do however be aware that if a standby is disconnected, WAL will continue to accumulate on the master until either the standby reconnects or the replication slot is dropped.

To enable repmgr to use replication slots, set the boolean parameter use_replication_slots in repmgr.conf:

use_replication_slots=1

Note that repmgr will fail with an error if this option is specified when working with PostgreSQL 9.3.

When cloning a standby, repmgr will automatically generate an appropriate slot name, which is stored in the repl_nodes table, and create the slot on the master:

repmgr=# SELECT * from repl_nodes ORDER BY id;
 id |  type   | upstream_node_id | cluster | name  |                 conninfo                 |   slot_name   | priority | active
----+---------+------------------+---------+-------+------------------------------------------+---------------+----------+--------
  1 | master  |                  | test    | node1 | host=localhost dbname=repmgr user=repmgr | repmgr_slot_1 |      100 | t
  2 | standby |                1 | test    | node2 | host=localhost dbname=repmgr user=repmgr | repmgr_slot_2 |      100 | t
  3 | standby |                1 | test    | node3 | host=localhost dbname=repmgr user=repmgr | repmgr_slot_3 |      100 | t

repmgr=# SELECT * FROM pg_replication_slots ;
   slot_name   | plugin | slot_type | datoid | database | active | active_pid | xmin | catalog_xmin | restart_lsn
---------------+--------+-----------+--------+----------+--------+------------+------+--------------+-------------
 repmgr_slot_3 |        | physical  |        |          | t      |      26060 |      |              | 0/50028F0
 repmgr_slot_2 |        | physical  |        |          | t      |      26079 |      |              | 0/50028F0
(2 rows)

Note that a slot name will be created by default for the master but not actually used unless the master is converted to a standby using e.g. repmgr standby switchover.

Be aware that when initially cloning a standby, you will need to ensure that all required WAL files remain available while the cloning is taking place. If using the default pg_basebackup method, we recommend setting pg_basebackup's --xlog-method parameter to stream like this:

pg_basebackup_options='--xlog-method=stream'

See the pg_basebackup documentation for details: http://www.postgresql.org/docs/current/static/app-pgbasebackup.html

Otherwise it's necessary to set wal_keep_segments to an appropriately high value.

Further information on replication slots in the PostgreSQL documentation: http://www.postgresql.org/docs/current/interactive/warm-standby.html#STREAMING-REPLICATION-SLOTS

If a master server fails or needs to be removed from the replication cluster, a new master server must be designated, to ensure the cluster continues working correctly. This can be done with repmgr standby promote, which promotes the standby on the current server to master

To demonstrate this, set up a replication cluster with a master and two attached standby servers so that the repl_nodes table looks like this:

repmgr=# SELECT * FROM repmgr_test.repl_nodes ORDER BY id;
 id |  type   | upstream_node_id | cluster | name  |                  conninfo                   | slot_name | priority | active
----+---------+------------------+---------+-------+---------------------------------------------+-----------+----------+--------
  1 | master  |                  | test    | node1 | host=repmgr_node1 dbname=repmgr user=repmgr |           |      100 | t
  2 | standby |                1 | test    | node2 | host=repmgr_node2 dbname=repmgr user=repmgr |           |      100 | t
  3 | standby |                1 | test    | node3 | host=repmgr_node3 dbname=repmgr user=repmgr |           |      100 | t
(3 rows)

Stop the current master with e.g.:

$ pg_ctl -D /path/to/node_1/data -m fast stop

At this point the replication cluster will be in a partially disabled state with both standbys accepting read-only connections while attempting to connect to the stopped master. Note that the repl_nodes table will not yet have been updated and will still show the master as active.

Promote the first standby with:

$ repmgr -f /etc/repmgr.conf standby promote

This will produce output similar to the following:

[2016-01-08 16:07:31] [ERROR] connection to database failed: could not connect to server: Connection refused
        Is the server running on host "repmgr_node1" (192.161.2.1) and accepting
        TCP/IP connections on port 5432?
could not connect to server: Connection refused
        Is the server running on host "repmgr_node1" (192.161.2.1) and accepting
        TCP/IP connections on port 5432?

[2016-01-08 16:07:31] [NOTICE] promoting standby
[2016-01-08 16:07:31] [NOTICE] promoting server using '/usr/bin/postgres/pg_ctl -D /path/to/node_2/data promote'
server promoting
[2016-01-08 16:07:33] [NOTICE] STANDBY PROMOTE successful

Note: the first [ERROR] is repmgr attempting to connect to the current master to verify that it has failed. If a valid master is found, repmgr will refuse to promote a standby.

The repl_nodes table will now look like this:

 id |  type   | upstream_node_id | cluster | name  |                  conninfo                   | slot_name | priority | active
----+---------+------------------+---------+-------+---------------------------------------------+-----------+----------+--------
  1 | master  |                  | test    | node1 | host=repmgr_node1 dbname=repmgr user=repmgr |           |      100 | f
  2 | master  |                  | test    | node2 | host=repmgr_node2 dbname=repmgr user=repmgr |           |      100 | t
  3 | standby |                1 | test    | node3 | host=repmgr_node3 dbname=repmgr user=repmgr |           |      100 | t
(3 rows)

The previous master has been marked as inactive, and node2's upstream_node_id has been cleared as it's now the "topmost" server in the replication cluster.

However the sole remaining standby is still trying to replicate from the failed master; repmgr standby follow must now be executed to rectify this situation.

Following the failure or removal of the replication cluster's existing master server, repmgr standby follow can be used to make 'orphaned' standbys follow the new master and catch up to its current state.

To demonstrate this, assuming a replication cluster in the same state as the end of the preceding section ("Promoting a standby server with repmgr"), execute this:

$ repmgr -f /etc/repmgr.conf -D /path/to/node_3/data/ -h repmgr_node2 -U repmgr -d repmgr standby follow
[2016-01-08 16:57:06] [NOTICE] restarting server using '/usr/bin/postgres/pg_ctl -D /path/to/node_3/data/ -w -m fast restart'
waiting for server to shut down.... done
server stopped
waiting for server to start.... done
server started

The standby is now replicating from the new master and repl_nodes has been updated to reflect this:

 id |  type   | upstream_node_id | cluster | name  |                  conninfo                   | slot_name | priority | active
----+---------+------------------+---------+-------+---------------------------------------------+-----------+----------+--------
  1 | master  |                  | test    | node1 | host=repmgr_node1 dbname=repmgr user=repmgr |           |      100 | f
  2 | master  |                  | test    | node2 | host=repmgr_node2 dbname=repmgr user=repmgr |           |      100 | t
  3 | standby |                2 | test    | node3 | host=repmgr_node3 dbname=repmgr user=repmgr |           |      100 | t
(3 rows)

Note that with cascading replication, repmgr standby follow can also be used to detach a standby from its current upstream server and follow another upstream server, including the master.

A typical use-case for replication is a combination of master and standby server, with the standby serving as a backup which can easily be activated in case of a problem with the master. Such an unplanned failover would normally be handled by promoting the standby, after which an appropriate action must be taken to restore the old master.

In some cases however it's desirable to promote the standby in a planned way, e.g. so maintenance can be performed on the master; this kind of switchover is supported by the repmgr standby switchover command.

repmgr standby switchover differs from other repmgr actions in that it also performs actions on another server, for which reason you must provide both passwordless SSH access and the path of repmgr.conf on that server.

NOTE repmgr standby switchover performs a relatively complex series of operations on two servers, and should therefore be performed after careful preparation and with adequate attention. In particular you should be confident that your network environment is stable and reliable.

We recommend running repmgr standby switchover at the most verbose logging level (--log-level DEBUG --verbose) and capturing all output to assist troubleshooting any problems.

Please also read carefully the list of caveats below.

To demonstrate switchover, we will assume a replication cluster running on PostgreSQL 9.5 or later with a master (node1) and a standby (node2); after the switchover node2 should become the master with node1 following it.

The switchover command must be run from the standby which is to be promoted, and in its simplest form looks like this:

repmgr -f /etc/repmgr.conf -C /etc/repmgr.conf standby switchover

-f /etc/repmgr.conf is, as usual the local repmgr node's configuration file. -C /etc/repmgr.conf is the path to the configuration file on the current master, which is required to execute repmgr remotely on that server; if it is not provided with -C, repmgr will check the same path as on the local server, as well as the normal default locations. repmgr will check this file can be found before performing any further actions.

$ repmgr -f /etc/repmgr.conf -C /etc/repmgr.conf standby switchover -v
[2016-01-27 16:38:33] [NOTICE] using configuration file "/etc/repmgr.conf"
[2016-01-27 16:38:33] [NOTICE] switching current node 2 to master server and demoting current master to standby...
[2016-01-27 16:38:34] [NOTICE] 5 files copied to /tmp/repmgr-node1-archive
[2016-01-27 16:38:34] [NOTICE] connection to database failed: FATAL:  the database system is shutting down

[2016-01-27 16:38:34] [NOTICE] current master has been stopped
[2016-01-27 16:38:34] [ERROR] connection to database failed: FATAL:  the database system is shutting down

[2016-01-27 16:38:34] [NOTICE] promoting standby
[2016-01-27 16:38:34] [NOTICE] promoting server using '/usr/local/bin/pg_ctl -D /var/lib/postgresql/9.5/node_2/data promote'
server promoting
[2016-01-27 16:38:36] [NOTICE] STANDBY PROMOTE successful
[2016-01-27 16:38:36] [NOTICE] Executing pg_rewind on old master server
[2016-01-27 16:38:36] [NOTICE] 5 files copied to /var/lib/postgresql/9.5/data
[2016-01-27 16:38:36] [NOTICE] restarting server using '/usr/local/bin/pg_ctl -w -D /var/lib/postgresql/9.5/node_1/data -m fast restart'
pg_ctl: PID file "/var/lib/postgresql/9.5/node_1/data/postmaster.pid" does not exist
Is server running?
starting server anyway
[2016-01-27 16:38:37] [NOTICE] node 1 is replicating in state "streaming"
[2016-01-27 16:38:37] [NOTICE] switchover was successful

Messages containing the line connection to database failed: FATAL: the database system is shutting down are not errors - repmgr is polling the old master database to make sure it has shut down correctly. repmgr will also archive any configuration files in the old master's data directory as they will otherwise be overwritten by pg_rewind; they are restored once the pg_rewind operation has completed.

The old master is now replicating as a standby from the new master and repl_nodes should have been updated to reflect this:

repmgr=# SELECT * from repl_nodes ORDER BY id;
 id |  type   | upstream_node_id | cluster | name  |                 conninfo                 | slot_name | priority | active
----+---------+------------------+---------+-------+------------------------------------------+-----------+----------+--------
  1 | standby |                2 | test    | node1 | host=localhost dbname=repmgr user=repmgr |           |      100 | t
  2 | master  |                  | test    | node2 | host=localhost dbname=repmgr user=repmgr |           |      100 | t
(2 rows)

• the functionality provided repmgr standby switchover is primarily aimed at a two-server master/standby replication cluster and currently does not support additional standbys.
• repmgr standby switchover is designed to use the pg_rewind utility, standard in 9.5 and later and available for seperately in 9.3 and 9.4 (see note below)
• pg_rewind requires that either wal_log_hints is enabled, or that data checksums were enabled when the cluster was initialized. See the pg_rewind documentation for details: http://www.postgresql.org/docs/current/static/app-pgrewind.html
• repmgrd should not be running when a switchover is carried out, otherwise the repmgrd may try and promote a standby by itself.
• Any other standbys attached to the old master will need to be manually instructed to point to the new master (e.g. with repmgr standby follow).

We hope to remove some of these restrictions in future versions of repmgr.

In order to efficiently reintegrate a demoted master into the replication cluster as a standby, it's necessary to resynchronise its data directory with that of the current master, as it's very likely that their timelines will have diverged slightly following the shutdown of the old master.

The utility pg_rewind provides an efficient way of doing this, however is not included in the core PostgreSQL distribution for versions 9.3 and 9.4. Hoever, pg_rewind is available separately for these versions and we strongly recommend its installation. To use it with versions 9.3 and 9.4, provide the command line option --pg_rewind, optionally with the path to the pg_rewind binary location if not installed in the PostgreSQL bin directory.

pg_rewind for versions 9.3 and 9.4 can be obtained from: https://github.com/vmware/pg_rewind

If pg_rewind is not available, as a fallback repmgr will use repmgr standby clone to resynchronise the old master's data directory using rsync. However, in order to ensure all files are synchronised, the entire data directory on both servers must be scanned, a process which can take some time on larger databases, in which case you should consider making a fresh standby clone.

To unregister a running standby, execute:

repmgr standby unregister -f /etc/repmgr.conf

This will remove the standby record from repmgr's internal metadata table (repl_nodes). A standby_unregister event notification will be recorded in the repl_events table.

Note that this command will not stop the server itself or remove it from the replication cluster.

If the standby is not running, the standby record must be manually removed from the repl_nodes table with e.g.:

DELETE FROM repmgr_test.repl_nodes WHERE id = 3;

Adjust schema and node ID accordingly. A future repmgr release will make it possible to unregister failed standbys.

repmgrd is a management and monitoring daemon which runs on standby nodes and which can automate actions such as failover and updating standbys to follow the new master.

To use repmgrd for automatic failover, the following repmgrd options must be set in repmgr.conf:

failover=automatic
promote_command='repmgr standby promote -f /etc/repmgr/repmgr.conf'
follow_command='repmgr standby follow -f /etc/repmgr/repmgr.conf'

(See repmgr.conf.sample for further repmgrd-specific settings).

Additionally, postgresql.conf must contain the following line:

shared_preload_libraries = 'repmgr_funcs'

When failover is set to automatic, upon detecting failure of the current master, repmgrd will execute one of promote_command or follow_command, depending on whether the current server is becoming the new master or needs to follow another server which has become the new master. Note that these commands can be any valid shell script which results in one of these actions happening, but we strongly recommend executing repmgr directly.

repmgrd can be started simply with e.g.:

repmgrd -f /etc/repmgr.conf --verbose > $HOME/repmgr/repmgr.log 2>&1

For permanent operation, we recommend using the options -d/--daemonize to detach the repmgrd process, and -p/--pid-file to write the process PID to a file.

Note that currently repmgrd is not required to run on the master server.

To demonstrate automatic failover, set up a 3-node replication cluster (one master and two standbys streaming directly from the master) so that the repl_nodes table looks like this:

repmgr=# SELECT * FROM repmgr_test.repl_nodes ORDER BY id;
 id |  type   | upstream_node_id | cluster | name  |                  conninfo                   | slot_name | priority | active
----+---------+------------------+---------+-------+---------------------------------------------+-----------+----------+--------
  1 | master  |                  | test    | node1 | host=repmgr_node1 dbname=repmgr user=repmgr |           |      100 | t
  2 | standby |                1 | test    | node2 | host=repmgr_node2 dbname=repmgr user=repmgr |           |      100 | t
  3 | standby |                1 | test    | node3 | host=repmgr_node3 dbname=repmgr user=repmgr |           |      100 | t
(3 rows)

Start repmgrd on each standby and verify that it's running by examining the log output, which at default log level will look like this:

[2016-01-05 13:15:40] [INFO] checking cluster configuration with schema 'repmgr_test'
[2016-01-05 13:15:40] [INFO] checking node 2 in cluster 'test'
[2016-01-05 13:15:40] [INFO] reloading configuration file and updating repmgr tables
[2016-01-05 13:15:40] [INFO] starting continuous standby node monitoring

Each repmgrd should also have noted its successful startup in the repl_events table:

repmgr=# SELECT * FROM repl_events WHERE event = 'repmgrd_start';
 node_id |     event     | successful |        event_timestamp        | details
---------+---------------+------------+-------------------------------+---------
       2 | repmgrd_start | t          | 2016-01-27 18:22:38.080231+09 |
       3 | repmgrd_start | t          | 2016-01-27 18:22:38.08756+09  |
(2 rows)

Now stop the current master server with e.g.:

pg_ctl -D /path/to/node1/data -m immediate stop

This will force the master node to shut down straight away, aborting all processes and transactions. This will cause a flurry of activity in the repmgrd log files as each repmgrd detects the failure of the master and a failover decision is made. Here extracts from the standby server promoted to new master:

[2016-01-06 18:32:58] [WARNING] connection to upstream has been lost, trying to recover... 15 seconds before failover decision
[2016-01-06 18:33:03] [WARNING] connection to upstream has been lost, trying to recover... 10 seconds before failover decision
[2016-01-06 18:33:08] [WARNING] connection to upstream has been lost, trying to recover... 5 seconds before failover decision
...
[2016-01-06 18:33:18] [NOTICE] this node is the best candidate to be the new master, promoting...
...
[2016-01-06 18:33:20] [NOTICE] STANDBY PROMOTE successful

and here from the standby server which is now following the new master:

[2016-01-06 18:32:58] [WARNING] connection to upstream has been lost, trying to recover... 15 seconds before failover decision
[2016-01-06 18:33:03] [WARNING] connection to upstream has been lost, trying to recover... 10 seconds before failover decision
[2016-01-06 18:33:08] [WARNING] connection to upstream has been lost, trying to recover... 5 seconds before failover decision
...
[2016-01-06 18:33:23] [NOTICE] node 2 is the best candidate for new master, attempting to follow...
[2016-01-06 18:33:23] [INFO] changing standby's master
...
[2016-01-06 18:33:25] [NOTICE] node 3 now following new upstream node 2

The repl_nodes table should have been updated to reflect the new situation, with the original master (node1) marked as inactive, and standby node3 now following the new master (node2):

repmgr=# SELECT * from repl_nodes ORDER BY id;
 id |  type   | upstream_node_id | cluster | name  |                 conninfo                 | slot_name | priority | active
----+---------+------------------+---------+-------+------------------------------------------+-----------+----------+--------
  1 | master  |                  | test    | node1 | host=localhost dbname=repmgr user=repmgr |           |      100 | f
  2 | master  |                  | test    | node2 | host=localhost dbname=repmgr user=repmgr |           |      100 | t
  3 | standby |                2 | test    | node3 | host=localhost dbname=repmgr user=repmgr |           |      100 | t
(3 rows)

The repl_events table will contain a summary of what happened to each server during the failover:

repmgr=# SELECT * from repmgr_test.repl_events where event_timestamp>='2016-01-06 18:30';
 node_id |          event           | successful |        event_timestamp        |                         details
---------+--------------------------+------------+-------------------------------+----------------------------------------------------------
       2 | standby_promote          | t          | 2016-01-06 18:33:20.061736+09 | node 2 was successfully promoted to master
       2 | repmgrd_failover_promote | t          | 2016-01-06 18:33:20.067132+09 | node 2 promoted to master; old master 1 marked as failed
       3 | repmgrd_failover_follow  | t          | 2016-01-06 18:33:25.331012+09 | node 3 now following new upstream node 2
(3 rows)

Note that currently repmgrd does not provide logfile rotation. To ensure the current logfile does not grow indefinitely, configure your system's logrotate to do this. Sample configuration to rotate logfiles weekly with retention for up to 52 weeks and rotation forced if a file grows beyond 100Mb:

/var/log/postgresql/repmgr-9.5.log {
    missingok
    compress
    rotate 52
    maxsize 100M
    weekly
    create 0600 postgres postgres
}

When repmgrd is running with the option -m/--monitoring-history, it will constantly write standby node status information to the repl_monitor table, providing a near-real time overview of replication status on all nodes in the cluster.

The view repl_status shows the most recent state for each node, e.g.:

repmgr=# SELECT * FROM repmgr_test.repl_status;
-[ RECORD 1 ]-------------+-----------------------------
primary_node              | 1
standby_node              | 2
standby_name              | node2
node_type                 | standby
active                    | t
last_monitor_time         | 2016-01-05 14:02:34.51713+09
last_wal_primary_location | 0/3012AF0
last_wal_standby_location | 0/3012AF0
replication_lag           | 0 bytes
replication_time_lag      | 00:00:03.463085
apply_lag                 | 0 bytes
communication_time_lag    | 00:00:00.955385

The interval in which monitoring history is written is controlled by the configuration parameter monitor_interval_secs; default is 2.

As this can generate a large amount of monitoring data in the repl_monitor table , it's advisable to regularly purge historical data with repmgr cluster cleanup; use the -k/--keep-history to specify how many day's worth of data should be retained.

Note that when a standby node is not streaming directly from its upstream node, i.e. recovering WAL from an archive, apply_lag will always appear as 0 bytes.

In a situation caused e.g. by a network interruption between two data centres, it's important to avoid a "split-brain" situation where both sides of the network assume they are the active segment and the side without an active master unilaterally promotes one of its standbys.

To prevent this situation happening, it's essential to ensure that one network segment has a "voting majority", so other segments will know they're in the minority and not attempt to promote a new master. Where an odd number of servers exists, this is not an issue. However, if each network has an even number of nodes, it's necessary to provide some way of ensuring a majority, which is where the witness server becomes useful.

This is not a fully-fledged standby node and is not integrated into replication, but it effectively represents the "casting vote" when deciding which network segment has a majority. A witness server can be set up using repmgr witness create (see below for details) and can run on a dedicated server or an existing node. Note that it only makes sense to create a witness server in conjunction with running repmgrd; the witness server will require its own repmgrd instance.

Cascading replication - where a standby can connect to an upstream node and not the master server itself - was introduced in PostgreSQL 9.2. repmgr and repmgrd support cascading replication by keeping track of the relationship between standby servers - each node record is stored with the node id of its upstream ("parent") server (except of course the master server).

In a failover situation where the master node fails and a top-level standby is promoted, a standby connected to another standby will not be affected and continue working as normal (even if the upstream standby it's connected to becomes the master node). If however the node's direct upstream fails, the "cascaded standby" will attempt to reconnect to that node's parent.

Each time repmgr or repmgrd perform a significant event, a record of that event is written into the repl_events table together with a timestamp, an indication of failure or success, and further details if appropriate. This is useful for gaining an overview of events affecting the replication cluster. However note that this table has advisory character and should be used in combination with the repmgr and PostgreSQL logs to obtain details of any events.

Example output after a master was registered and a standby cloned and registered:

repmgr=# SELECT * from repmgr_test.repl_events ;
 node_id |      event       | successful |        event_timestamp        |                                       details
---------+------------------+------------+-------------------------------+-------------------------------------------------------------------------------------
       1 | master_register  | t          | 2016-01-08 15:04:39.781733+09 |
       2 | standby_clone    | t          | 2016-01-08 15:04:49.530001+09 | Cloned from host 'repmgr_node1', port 5432; backup method: pg_basebackup; --force: N
       2 | standby_register | t          | 2016-01-08 15:04:50.621292+09 |
(3 rows)

Additionally, event notifications can be passed to a user-defined program or script which can take further action, e.g. send email notifications. This is done by setting the event_notification_command parameter in repmgr.conf.

This parameter accepts the following format placeholders:

 %n - node ID
 %e - event type
 %s - success (1 or 0)
 %t - timestamp
 %d - details

The values provided for "%t" and "%d" will probably contain spaces, so should be quoted in the provided command configuration, e.g.:

event_notification_command='/path/to/some/script %n %e %s "%t" "%d"'

By default, all notifications will be passed; the notification types can be filtered to explicitly named ones:

event_notifications=master_register,standby_register,witness_create

The following event types are available:

• master_register
• standby_register
• standby_unregister
• standby_clone
• standby_promote
• standby_follow
• standby_switchover
• witness_create
• witness_create
• repmgrd_start
• repmgrd_shutdown
• repmgrd_failover_promote
• repmgrd_failover_follow

Note that under some circumstances (e.g. no replication cluster master could be located), it will not be possible to write an entry into the repl_events table, in which case event_notification_command can serve as a fallback.

repmgr is updated regularly with point releases (e.g. 3.0.2 to 3.0.3) containing bugfixes and other minor improvements. Any substantial new functionality will be included in a feature release (e.g. 3.0.x to 3.1.x).

In general repmgr can be upgraded as-is without any further action required, however feature releases may require the repmgr database to be upgraded. An SQL script will be provided - please check the release notes for details.

For some command line and most configuration file parameters, repmgr falls back to default values if values for these are not explicitly provided.

The file repmgr.conf.sample documents the default value of configuration parameters if one is set. Of particular note is the log level, which defaults to NOTICE; particularly when using repmgr from the command line it may be useful to set this to a higher level with -L/--log-level. e.g. to INFO.

Execute repmgr --help to see the default values for various command line parameters, particularly database connection parameters.

See the section Configuration above for information on how the configuration file is located if -f/--config-file is not supplied.

The repmgr command line tool accepts commands for specific servers in the replication in the format "server_type action", or for the entire replication cluster in the format "cluster action". Each command is described below.

In general, each command needs to be provided with the path to repmgr.conf, which contains connection details for the local database.

• master register

  Registers a master in a cluster. This command needs to be executed before any standby nodes are registered.

  primary register can be used as an alias for master register.

• standby register

  Registers a standby with repmgr. This command needs to be executed to enable promote/follow operations and to allow repmgrd to work with the node. An existing standby can be registered using this command.

• standby unregister

  Unregisters a standby with repmgr. This command does not affect the actual replication, just removes the standby's entry from the repl_nodes table.

• standby clone [node to be cloned]

  Clones a new standby node from the data directory of the master (or an upstream cascading standby) using pg_basebackup or rsync. Additionally it will create the recovery.conf file required to start the server as a standby. This command does not require repmgr.conf to be provided, but does require connection details of the master or upstream server as command line parameters.

  Provide the -D/--data-dir option to specify the destination data directory; if not, the same directory path as on the source server will be used. By default, pg_basebackup will be used to copy data from the master or upstream node but this can only be used for bootstrapping new installations. To update an existing but 'stale' data directory (for example belonging to a failed master), rsync must be used by specifying --rsync-only. In this case, password-less SSH connections between servers are required.

• standby promote

  Promotes a standby to a master if the current master has failed. This command requires a valid repmgr.conf file for the standby, either specified explicitly with -f/--config-file or located in the current working directory; no additional arguments are required.

  If the standby promotion succeeds, the server will not need to be restarted. However any other standbys will need to follow the new server, by using standby follow (see below); if repmgrd is active, it will handle this.

  This command will not function if the current master is still running.

• standby switchover

  Promotes a standby to master and demotes the existing master to a standby. This command must be run on the standby to be promoted, and requires a password-less SSH connection to the current master. Additionally the location of the master's repmgr.conf file must be provided with -C/--remote-config-file.

  repmgrd should not be active if a switchover is attempted. This restriction may be lifted in a later version.

• standby follow

  Attaches the standby to a new master. This command requires a valid repmgr.conf file for the standby, either specified explicitly with -f/--config-file or located in the current working directory; no additional arguments are required.

  This command will force a restart of the standby server. It can only be used to attach a standby to a new master node.

• witness create

  Creates a witness server as a separate PostgreSQL instance. This instance can be on a separate server or a server running an existing node. The witness server contain a copy of the repmgr metadata tables but will not be set up as a standby; instead it will update its metadata copy each time a failover occurs.

  Note that it only makes sense to create a witness server if repmgrd is in use; see section "Using a witness server" above.

  This command requires a repmgr.conf file containing a valid conninfo string for the server to be created, as well as the other minimum required parameters detailed in the section repmgr configuration file above.

  By default the witness server will use port 5499 to facilitate easier setup on a server running an existing node. To use a different port, supply this explicitly in the repmgr.conf conninfo string.

  This command also requires the location of the witness server's data directory to be provided (-D/--datadir) as well as valid connection parameters for the master server.

  By default this command will create a superuser and a repmgr user. The repmgr user name will be extracted from the conninfo string in repmgr.conf.

• cluster show

  Displays information about each active node in the replication cluster. This command polls each registered server and shows its role (master / standby / witness) or FAILED if the node doesn't respond. It polls each server directly and can be run on any node in the cluster; this is also useful when analyzing connectivity from a particular node.

  This command requires a valid repmgr.conf file to be provided; no additional arguments are required.

  Example:

    $ repmgr -f /etc/repmgr.conf cluster show
  
    Role      | Name  | Upstream | Connection String
    ----------+-------|----------|--------------------------------------------
    * master  | node1 |          | host=repmgr_node1 dbname=repmgr user=repmgr
      standby | node2 | node1    | host=repmgr_node1 dbname=repmgr user=repmgr
      standby | node3 | node2    | host=repmgr_node1 dbname=repmgr user=repmgr
  

• cluster cleanup

  Purges monitoring history from the repl_monitor table to prevent excessive table growth. Use the -k/--keep-history to specify the number of days of monitoring history to retain. This command can be used manually or as a cronjob.

  This command requires a valid repmgr.conf file for the node on which it is executed, either specified explicitly with -f/--config-file or located in the current working directory; no additional arguments are required.

repmgr or repmgrd will return one of the following error codes on program exit:

• SUCCESS (0) Program ran successfully.
• ERR_BAD_CONFIG (1) Configuration file could not be parsed or was invalid
• ERR_BAD_RSYNC (2) An rsync call made by the program returned an error
• ERR_NO_RESTART (4) An attempt to restart a PostgreSQL instance failed
• ERR_DB_CON (6) Error when trying to connect to a database
• ERR_DB_QUERY (7) Error while executing a database query
• ERR_PROMOTED (8) Exiting program because the node has been promoted to master
• ERR_BAD_PASSWORD (9) Password used to connect to a database was rejected
• ERR_STR_OVERFLOW (10) String overflow error
• ERR_FAILOVER_FAIL (11) Error encountered during failover (repmgrd only)
• ERR_BAD_SSH (12) Error when connecting to remote host via SSH
• ERR_SYS_FAILURE (13) Error when forking (repmgrd only)
• ERR_BAD_BASEBACKUP (14) Error when executing pg_basebackup
• ERR_MONITORING_FAIL (16) Unrecoverable error encountered during monitoring (repmgrd only)

2ndQuadrant provides 24x7 production support for repmgr, including configuration assistance, installation verification and training for running a robust replication cluster. For further details see:

• http://2ndquadrant.com/en/support/

There is a mailing list/forum to discuss contributions or issues:

• http://groups.google.com/group/repmgr

The IRC channel #repmgr is registered with freenode.

Please report bugs and other issues to:

• https://github.com/2ndQuadrant/repmgr

Further information is available at http://www.repmgr.org/

We'd love to hear from you about how you use repmgr. Case studies and news are always welcome. Send us an email at info@2ndQuadrant.com, or send a postcard to

repmgr
c/o 2ndQuadrant
7200 The Quorum
Oxford Business Park North
Oxford
OX4 2JZ
United Kingdom

Thanks from the repmgr core team.

• Ian Barwick
• Jaime Casanova
• Abhijit Menon-Sen
• Simon Riggs
• Cedric Villemain

• http://blog.2ndquadrant.com/managing-useful-clusters-repmgr/
• http://blog.2ndquadrant.com/easier_postgresql_90_clusters/