Database settings

Tier: Free, Premium, Ultimate Offering: Self-managed

GitLab supports only the PostgreSQL database management system.

Thus you have two options for database servers to use with a Linux package installation:

  • Use the packaged PostgreSQL server included with the Linux package installation (no configuration required, recommended).
  • Use an external PostgreSQL server.

Using the PostgreSQL database service shipped with the Linux package

Reconfigure and PostgreSQL restarts

Linux package installations normally restart any service on reconfigure if configuration settings for that service were changed in the gitlab.rb file. PostgreSQL is unique in that some of its settings take effect with a reload (HUP), while others require PostgreSQL to be restarted. Because administrators frequently want more control over exactly when PostgreSQL is restarted, Linux package installations are configured to do a reload of PostgreSQL on reconfigure, and not a restart. This means that if you modify any PostgreSQL setting that requires a restart, you will need to restart PostgreSQL manually after you reconfigure.

The GitLab config template identifies which PostgreSQL settings require a restart and which require only a reload. You can also run a query against your database to determine if any individual setting requires a restart. Start a database console with sudo gitlab-psql, then replace <setting name> in the following query with the setting you are changing: 

SELECT name,setting FROM pg_settings WHERE context = 'postmaster' AND name = '<setting name>';

If changing the setting will require a restart, the query will return the name of the setting and the current value of that setting in the running PostgreSQL instance.

Automatic restart when the PostgreSQL version changes

By default, Linux package installations automatically restart PostgreSQL when the underlying version changes, as suggested by the upstream documentation. This behavior can be controlled using the auto_restart_on_version_change setting available for postgresql and geo-postgresql.

To disable automatic restarts when the PostgreSQL version changes:

  1. Edit /etc/gitlab/gitlab.rb and add the following line: 

    # For PostgreSQL/Patroni
postgresql['auto_restart_on_version_change'] = false

# For Geo PostgreSQL
geo_postgresql['auto_restart_on_version_change'] = false

  2. Reconfigure GitLab: 

    sudo gitlab-ctl reconfigure
note
It is highly recommended to restart PostgreSQL when the underlying version changes, to avoid errors like the one related to loading necessary libraries.

Configuring SSL

Linux package installations automatically enable SSL on the PostgreSQL server, but it will accept both encrypted and unencrypted connections by default. Enforcing SSL requires using the hostssl configuration in pg_hba.conf. For more details, see the pg_hba.conf documentation.

SSL support depends on the following files:

  • The public SSL certificate for the database (server.crt).
  • The corresponding private key for the SSL certificate (server.key).
  • A root certificate bundle that validates the server’s certificate (root.crt). By default, Linux package installations use the embedded certificate bundle in /opt/gitlab/embedded/ssl/certs/cacert.pem. This isn’t required for self-signed certificates.

A 10-year self-signed certificate and private key are generated by a Linux package installation for use. If you’d prefer to use a CA-signed certificate or replace this with your own self-signed certificate, use the following steps.

Note that the location of these files can be configurable, but the private key must be readable by the gitlab-psql user. Linux package installations manage the permissions of the files for you, but if the paths are customized, you must ensure that the gitlab-psql can access the directory in the files are placed in.

For more details, see the PostgreSQL documentation.

Note that server.crt and server.key may be different from the default SSL certificates used to access GitLab. For example, suppose the external hostname of your database is database.example.com, and your external GitLab hostname is gitlab.example.com. You will either need a wildcard certificate for *.example.com or two different SSL certificates.

The ssl_cert_file, ssl_key_file, and ssl_ca_file files direct PostgreSQL to where on the filesystem to find the certificate, key, and bundle. These changes are applied to postgresql.conf. The directives internal_certificate and internal_key are used to populate the contents of these files. The contents can be added directly or loaded from file as shown in the following example.

After you have these files, enable SSL:

  1. Edit /etc/gitlab/gitlab.rb: 

    postgresql['ssl_cert_file'] = '/custom/path/to/server.crt'
postgresql['ssl_key_file'] = '/custom/path/to/server.key'
postgresql['ssl_ca_file'] = '/custom/path/to/bundle.pem'
postgresql['internal_certificate'] = File.read('/custom/path/to/server.crt')
postgresql['internal_key'] = File.read('/custom/path/to/server.key')

    Relative paths will be rooted in the PostgreSQL data directory (/var/opt/gitlab/postgresql/data by default).

  2. Reconfigure GitLab to apply the configuration changes.

  3. Restart PostgreSQL for the changes to take effect: 

    gitlab-ctl restart postgresql

    If PostgreSQL fails to start, check the logs (for example, /var/log/gitlab/postgresql/current) for more details.

Require SSL

  1. Add the following to /etc/gitlab/gitlab.rb: 

    gitlab_rails['db_sslmode'] = 'require'

  2. Reconfigure GitLab to apply the configuration changes.

  3. Restart PostgreSQL for the changes to take effect: 

    gitlab-ctl restart postgresql

    If PostgreSQL fails to start, check the logs (for example, /var/log/gitlab/postgresql/current) for more details.

Disabling SSL

  1. Add the following to /etc/gitlab/gitlab.rb: 

    postgresql['ssl'] = 'off'

  2. Reconfigure GitLab to apply the configuration changes.

  3. Restart PostgreSQL for the changes to take effect: 

    gitlab-ctl restart postgresql

    If PostgreSQL fails to start, check the logs (for example, /var/log/gitlab/postgresql/current) for more details.

Verifying that SSL is being used

To determine whether SSL is being used by clients, you can run: 

sudo gitlab-rails dbconsole --database main

At startup, you should see a banner as the following: 

psql (13.14)
SSL connection (protocol: TLSv1.2, cipher: ECDHE-RSA-AES256-GCM-SHA384, bits: 256, compression: on)
Type "help" for help.

To determine if clients are using SSL, issue this SQL query: 

SELECT * FROM pg_stat_ssl;

For example: 

gitlabhq_production=> select * from pg_stat_ssl;
 pid  | ssl | version |         cipher         | bits | compression |  clientdn
------+-----+---------+------------------------+------+-------------+------------
  384 | f   |         |                        |      |             |
  386 | f   |         |                        |      |             |
  998 | t   | TLSv1.3 | TLS_AES_256_GCM_SHA384 |  256 | f           | /CN=gitlab
  933 | f   |         |                        |      |             |
 1003 | t   | TLSv1.3 | TLS_AES_256_GCM_SHA384 |  256 | f           | /CN=gitlab
 1016 | t   | TLSv1.3 | TLS_AES_256_GCM_SHA384 |  256 | f           | /CN=gitlab
 1022 | t   | TLSv1.3 | TLS_AES_256_GCM_SHA384 |  256 | f           | /CN=gitlab
 1211 | t   | TLSv1.3 | TLS_AES_256_GCM_SHA384 |  256 | f           | /CN=gitlab
 1214 | t   | TLSv1.3 | TLS_AES_256_GCM_SHA384 |  256 | f           | /CN=gitlab
 1213 | t   | TLSv1.3 | TLS_AES_256_GCM_SHA384 |  256 | f           | /CN=gitlab
 1215 | t   | TLSv1.3 | TLS_AES_256_GCM_SHA384 |  256 | f           | /CN=gitlab
 1252 | t   | TLSv1.3 | TLS_AES_256_GCM_SHA384 |  256 | f           |
 1280 | t   | TLSv1.3 | TLS_AES_256_GCM_SHA384 |  256 | f           | /CN=gitlab
  382 | f   |         |                        |      |             |
  381 | f   |         |                        |      |             |
  383 | f   |         |                        |      |             |
(16 rows)
  1. Rows that have t listed under the ssl column are enabled.
  2. Rows that have a value in the clientdn are using the cert authentication method

Configure SSL client authentication

Client SSL certificates can be used to authenticate to the database server. Creating the certificates is beyond the scope of omnibus-gitlab. But users who have an existing SSL certificate management solution can use this.

Configure the database server
  1. Create a certificate and key for the server, the common name should equal the DNS name of the server
  2. Copy the server certificate, key, and CA file to the PostgreSQL server, and ensure the permissions are correct
    1. The certificate should be owned by the database user (default: gitlab-psql)
    2. The key file should be owned by the database user, and its permissions should be 0400
    3. The CA file should be owned by the database user, and its permissions should be 0400
    note
    Don’t use the file names server.crt or server.key for these files. These file names are reserved for the internal use of omnibus-gitlab.

  3. Ensure the following is set in gitlab.rb: 

    postgresql['ssl_cert_file'] = 'PATH_TO_CERTIFICATE'
postgresql['ssl_key_file'] = 'PATH_TO_KEY_FILE'
postgresql['ssl_ca_file'] = 'PATH_TO_CA_FILE'
postgresql['listen_address'] = 'IP_ADDRESS'
postgresql['cert_auth_addresses'] = {
'IP_ADDRESS' => {
  'database' => 'gitlabhq_production',
  'user' => 'gitlab'
}

    Set listen_address as the IP address of the server that the clients will use to connect to the database. Ensure cert_auth_addresses contains a list of IP addresses and the databases and users that are allowed to connect to the database. You can use CIDR notation when specifying the key for cert_auth_addresses to incorporate an IP address range.

  4. Run gitlab-ctl reconfigure, and then gitlab-ctl restart postgresql for the new settings to take effect.

Configure the Rails client

For the rails client to connect to the server, you will need a certificate and key with the commonName set to gitlab, which is signed by a certificate authority trusted in the CA file specified in ssl_ca_file on the database server.

  1. Configure gitlab.rb 

    gitlab_rails['db_host'] = 'IP_ADDRESS_OR_HOSTNAME_OF_DATABASE_SERVER'
gitlab_rails['db_sslcert'] = 'PATH_TO_CERTIFICATE_FILE'
gitlab_rails['db_sslkey'] = 'PATH_TO_KEY_FILE'
gitlab_rails['db_rootcert'] = 'PATH_TO_CA_FILE'
  2. Run gitlab-ctl reconfigure for the rails client to use the new settings
  3. Follow the steps in Verifying that SSL is being used to ensure the authentication is working.

Configure packaged PostgreSQL server to listen on TCP/IP

The packaged PostgreSQL server can be configured to listen for TCP/IP connections, with the caveat that some non-critical scripts expect UNIX sockets and may misbehave.

To configure the use of TCP/IP for the database service, make changes to both the postgresql and gitlab_rails sections of gitlab.rb.

Configure PostgreSQL block

The following settings are affected in the postgresql block:

  • listen_address: Controls the address on which PostgreSQL will listen.
  • port: Controls the port on which PostgreSQL listens to. The default is 5432.
  • md5_auth_cidr_addresses: A list of CIDR address blocks that are allowed to connect to the server, after authentication with a password.
  • trust_auth_cidr_addresses: A list of CIDR address blocks that are allowed to connect to the server, without authentication of any kind. You should only set this setting to allow connections from nodes that need to connect, such as GitLab Rails or Sidekiq. This includes local connections when deployed on the same node or from components such as Postgres Exporter (127.0.0.1/32).
  • sql_user: Controls the expected username for MD5 authentication. This defaults to gitlab, and isn’t a required setting.
  • sql_user_password: Sets the password that PostgreSQL will accept for MD5 authentication.

  1. Edit /etc/gitlab/gitlab.rb: 

    postgresql['listen_address'] = '0.0.0.0'
postgresql['port'] = 5432
postgresql['md5_auth_cidr_addresses'] = %w()
postgresql['trust_auth_cidr_addresses'] = %w(127.0.0.1/24)
postgresql['sql_user'] = "gitlab"

##! SQL_USER_PASSWORD_HASH can be generated using the command `gitlab-ctl pg-password-md5 'gitlab'`,
##! where 'gitlab' (single-quoted to avoid shell interpolation) is the name of the SQL user that connects to GitLab.
##! You will be prompted for a password which other clients will use to authenticate with database, such as `securesqlpassword` in the below section.
postgresql['sql_user_password'] = "SQL_USER_PASSWORD_HASH"

# force ssl on all connections defined in trust_auth_cidr_addresses and md5_auth_cidr_addresses
postgresql['hostssl'] = true

  2. Reconfigure GitLab and restart PostrgreSQL: 

    sudo gitlab-ctl reconfigure
sudo gitlab-ctl restart postgresql

Any client or GitLab service which will connect over the network will need to provide the values of sql_user for the username, and password provided to the configuration when connecting to the PostgreSQL server. They must also be in the network block provided to md5_auth_cidr_addresses

Configure GitLab Rails block

To configure the gitlab-rails application to connect to the PostgreSQL database over the network, several settings must be configured:

  • db_host: Needs to be set to the IP address of the database server. If this is on the same instance as the PostgreSQL service, this can be 127.0.0.1 and will not require password authentication.
  • db_port: Sets the port on the PostgreSQL server to connect to, and must be set if db_host is set.
  • db_username: Configures the username with which to connect to PostgreSQL. This defaults to gitlab.
  • db_password: Must be provided if connecting to PostgreSQL over TCP/IP, and from an instance in the postgresql['md5_auth_cidr_addresses'] block from settings above. This is not required if you are connecting to 127.0.0.1 and have configured postgresql['trust_auth_cidr_addresses'] to include it.

  1. Edit /etc/gitlab/gitlab.rb: 

    gitlab_rails['db_host'] = '127.0.0.1'
gitlab_rails['db_port'] = 5432
gitlab_rails['db_username'] = "gitlab"
gitlab_rails['db_password'] = "securesqlpassword"

  2. Reconfigure GitLab and restart PostrgreSQL: 

    sudo gitlab-ctl reconfigure
sudo gitlab-ctl restart postgresql

Apply and restart services

After making the previous changes, an administrator should run gitlab-ctl reconfigure. If you experience any issues in regards to the service not listening on TCP, try directly restarting the service with gitlab-ctl restart postgresql.

Some included scripts of the Linux package (such as gitlab-psql) expect the connections to PostgreSQL to be handled over the UNIX socket, and may not function properly. You can enable TCP/IP without disabling UNIX sockets.

To test access from other clients, you can run: 

sudo gitlab-rails dbconsole --database main

Enabling PostgreSQL WAL (Write Ahead Log) Archiving

By default, WAL archiving of the packaged PostgreSQL isn’t enabled. Consider the following when seeking to enable WAL archiving:

  • The WAL level needs to be ‘replica’ or higher (9.6+ options are minimal, replica, or logical)
  • Increasing the WAL level will increase the amount of storage consumed in regular operations

To enable WAL Archiving:

  1. Edit /etc/gitlab/gitlab.rb: 

    # Replication settings
postgresql['sql_replication_user'] = "gitlab_replicator"
postgresql['wal_level'] = "replica"
    ...
    ...
# Backup/Archive settings
postgresql['archive_mode'] = "on"
postgresql['archive_command'] = "/your/wal/archiver/here"
postgresql['archive_timeout'] = "60"

  2. Reconfigure GitLab for the changes to take effect. This will result in a database restart.

Store PostgreSQL data in a different directory

By default, everything is stored under /var/opt/gitlab/postgresql, controlled by the postgresql['dir'] attribute.

This consists of:

  • The database socket will be /var/opt/gitlab/postgresql/.s.PGSQL.5432. This is controlled by postgresql['unix_socket_directory'].
  • The gitlab-psql system user will have its HOME directory set to this. This is controlled by postgresql['home'].
  • The actual data will be stored in /var/opt/gitlab/postgresql/data.

To change the location of the PostgreSQL data

caution
If you have an existing database, you need to move the data to the new location first.
caution
This is an intrusive operation. It cannot be done without downtime on an existing installation
  1. If this is an existing installation, stop GitLab: gitlab-ctl stop.
  2. Update postgresql['dir'] to the desired location.
  3. Run gitlab-ctl reconfigure.
  4. Start GitLab gitlab-ctl start.

Upgrade packaged PostgreSQL server

The Linux package provides the gitlab-ctl pg-upgrade command to update the packaged PostgreSQL server to a later version (if one is included in the package). This updates PostgreSQL to the default shipped version during package upgrades, unless specifically opted out.

Before upgrading GitLab to a newer version, refer to the version-specific changes of the Linux package to see either:

  • When a database version has changed.
  • When an upgrade is warranted.
caution
Before upgrading, it’s important that you fully read this section before running any commands. For single-node installations, this upgrade needs downtime, as the database must be down while the upgrade is being performed. The length of time depends on the size of your database.
note
If you encounter any problems during the upgrade, raise an issue with a full description at the omnibus-gitlab issue tracker.

To upgrade the PostgreSQL version, be sure that:

  • You’re running the latest version of GitLab that supports your current version of PostgreSQL.
  • If you recently upgraded, you ran sudo gitlab-ctl reconfigure successfully before you proceed.

  • You have sufficient disk space for two copies of your database. Do not attempt to upgrade unless you have enough free space available.

    • Check your database size using sudo du -sh /var/opt/gitlab/postgresql/data (or update your database path).
    • Check the space available using sudo df -h. If the partition where the database resides doesn’t have enough space, pass the argument --tmp-dir $DIR to the command. The upgrade task includes an available disk space check and aborts the upgrade if the requirements aren’t met.

After you confirm that the above checklist is satisfied, you can proceed with the upgrade: 

sudo gitlab-ctl pg-upgrade

To upgrade to a specific PostgreSQL version, use the -V flag to append the version. For example, to upgrade to PostgreSQL 14: 

sudo gitlab-ctl pg-upgrade -V 14
note
pg-upgrade can take arguments; for example, you can set the timeout for the execution of the underlying commands (--timeout=1d2h3m4s5ms). Run gitlab-ctl pg-upgrade -h to see the full list.

gitlab-ctl pg-upgrade performs the following steps:

  1. Checks to ensure the database is in a known good state.
  2. Checks if there’s enough free disk space and abort otherwise. You can skip this by appending the --skip-disk-check flag.
  3. Shuts down the existing database and any unnecessary services, and enables GitLab to deploy page.
  4. Changes the symlinks in /opt/gitlab/embedded/bin/ for PostgreSQL to point to the newer version of the database.
  5. Creates a new directory containing a new, empty database with a locale matching the existing database.
  6. Uses the pg_upgrade tool to copy the data from the old database to the new database.
  7. Moves the old database out of the way.
  8. Moves the new database to the expected location.
  9. Calls sudo gitlab-ctl reconfigure to do the required configuration changes and starts the new database server.
  10. Runs ANALYZE to generate database statistics.
  11. Starts the remaining services and removes the deploy page.
  12. If any errors are detected during this process, it reverts to the old version of the database.

After the upgrade is complete, verify that everything is working as expected.

If there was an error in the output while running the ANALYZE step, your upgrade will still be working but will have poor database performance until the database statistics are generated. Use gitlab-psql to determine whether ANALYZE should be run manually: 

sudo gitlab-psql -c "SELECT relname, last_analyze, last_autoanalyze FROM pg_stat_user_tables WHERE last_analyze IS NULL AND last_autoanalyze IS NULL;"

You can run ANALYZE manually if the query above returned any rows: 

sudo gitlab-psql -c 'SET statement_timeout = 0; ANALYZE VERBOSE;'

After you have verified that your GitLab instance is running correctly, you can clean up the old database files: 

sudo rm -rf /var/opt/gitlab/postgresql/data.<old_version>
sudo rm -f /var/opt/gitlab/postgresql-version.old

You can find details of PostgreSQL versions shipped with various GitLab versions in PostgreSQL versions shipped with the Linux package.

Opt out of automatic PostgreSQL upgrades

To opt out of automatic PostgreSQL upgrades during GitLab package upgrades, run: 

sudo touch /etc/gitlab/disable-postgresql-upgrade

Revert packaged PostgreSQL server to the previous version

caution
This operation will revert your current database, including its data, to its state before your last upgrade. Be sure to create a backup before attempting to downgrade your packaged PostgreSQL database.

On GitLab versions which ship multiple PostgreSQL versions, users can downgrade an already upgraded PostgreSQL version to the earlier version using the gitlab-ctl revert-pg-upgrade command. This command also supports the -V flag to specify a target version for scenarios where more than two PostgreSQL versions are shipped in the package (for example: GitLab 12.8 where PostgreSQL 9.6.x, 10.x, and 11.x are shipped).

To specify a target PostgreSQL version of 12: 

gitlab-ctl revert-pg-upgrade -V 12

If the target version is not specified, it will use the version in /var/opt/gitlab/postgresql-version.old if available. Otherwise, it falls back to the default version shipped with GitLab.

On other GitLab versions that ship only one PostgreSQL version, you can’t downgrade your PostgreSQL version. You must downgrade GitLab to an older version for this.

Configuring multiple database connections

History

In GitLab 16.0, GitLab defaults to using two database connections that point to the same PostgreSQL database.

Before upgrading to GitLab 16.0, check that the PostgreSQL max_connections setting is high enough so that more than 50% of available connections show as being unused. For example, if max_connections is set to 100 and you see 75 connections in use, you must increase max_connections to at least 150 before upgrading because after upgrading, the in-use connections will double to 150.

You can verify this by running the following Rake task: 

sudo gitlab-rake gitlab:db:decomposition:connection_status

If the task indicates that max_connections is high enough, then you can proceed with the upgrade.

Using a non-packaged PostgreSQL database management server

By default, GitLab is configured to use the PostgreSQL server that’s included in the Linux package. You can also reconfigure it to use an external instance of PostgreSQL.

caution
If you are using a non-packaged PostgreSQL server, you need to make sure that PostgreSQL is set up according to the database requirements document.

  1. Edit /etc/gitlab/gitlab.rb: 

    # Disable the built-in Postgres
postgresql['enable'] = false

# Fill in the connection details for database.yml
gitlab_rails['db_adapter'] = 'postgresql'
gitlab_rails['db_encoding'] = 'utf8'
gitlab_rails['db_host'] = '127.0.0.1'
gitlab_rails['db_port'] = 5432
gitlab_rails['db_username'] = 'USERNAME'
gitlab_rails['db_password'] = 'PASSWORD'

    Don’t forget to remove the # comment characters at the beginning of these lines.

    Note that:

    • /etc/gitlab/gitlab.rb should have file permissions 0600 because it contains plain-text passwords.

    • PostgreSQL allows listening on multiple addresses

      If you use multiple addresses in gitlab_rails['db_host'], comma-separated, the first address in the list will be used for the connection.

  2. Reconfigure GitLab for the changes to take effect.

  3. Seed the database.

UNIX socket configuration for non-packaged PostgreSQL

If you want to use your system’s PostgreSQL server (installed on the same system as GitLab) instead of the one bundled with GitLab, you can do so by using a UNIX socket:

  1. Edit /etc/gitlab/gitlab.rb: 

    # Disable the built-in Postgres
postgresql['enable'] = false

# Fill in the connection details for database.yml
gitlab_rails['db_adapter'] = 'postgresql'
gitlab_rails['db_encoding'] = 'utf8'
# The path where the socket lives
gitlab_rails['db_host'] = '/var/run/postgresql/'

  2. Reconfigure GitLab for the changes to take effect: 

    sudo gitlab-ctl-reconfigure

Configuring SSL

Require SSL

  1. Add the following to /etc/gitlab/gitlab.rb: 

    gitlab_rails['db_sslmode'] = 'require'

  2. Reconfigure GitLab to apply the configuration changes.

  3. Restart PostgreSQL for the changes to take effect: 

    gitlab-ctl restart postgresql

    If PostgreSQL fails to start, check the logs (for example, /var/log/gitlab/postgresql/current) for more details.

Require SSL and verify server certificate against CA bundle

PostgreSQL can be configured to require SSL and verify the server certificate against a CA bundle to prevent spoofing. The CA bundle that’s specified in gitlab_rails['db_sslrootcert'] must contain both the root and intermediate certificates.

  1. Add the following to /etc/gitlab/gitlab.rb: 

    gitlab_rails['db_sslmode'] = "verify-full"
gitlab_rails['db_sslrootcert'] = "<full_path_to_your_ca-bundle.pem>"

    If you are using Amazon RDS for your PostgreSQL server, ensure you download and use the combined CA bundle for gitlab_rails['db_sslrootcert']. More information on this can be found in the using SSL/TLS to Encrypt a Connection to a DB Instance article on AWS.

  2. Reconfigure GitLab to apply the configuration changes.

  3. Restart PostgreSQL for the changes to take effect: 

    gitlab-ctl restart postgresql

    If PostgreSQL fails to start, check the logs (for example, /var/log/gitlab/postgresql/current) for more details.

Backup and restore a non-packaged PostgreSQL database

When using the backup and restore commands, GitLab will attempt to use the packaged pg_dump command to create a database backup file and the packaged psql command to restore a backup. This will only work if they are the correct versions. Check the versions of the packaged pg_dump and psql: 

/opt/gitlab/embedded/bin/pg_dump --version
/opt/gitlab/embedded/bin/psql --version

If these versions are different from your non-packaged external PostgreSQL, you may encounter the following error output when attempting to run the backup command. 

Dumping PostgreSQL database gitlabhq_production ... pg_dump: error: server version: 13.3; pg_dump version: 12.6
pg_dump: error: aborting because of server version mismatch

In this example, the error occurs on GitLab 14.1 when using PostgreSQL version 13.3, instead of the default shipped PostgreSQL version of 12.6.

In this case, you will need to install tools that match your database version and then follow the steps below. There are multiple ways to install PostgreSQL client tools. See https://www.postgresql.org/download/ for options.

Once the correct psql and pg_dump tools are available on your system, follow these steps, using the correct path to the location you installed the new tools:

  1. Add symbolic links to the non-packaged versions: 

    ln -s /path/to/new/pg_dump /path/to/new/psql /opt/gitlab/bin/

  2. Check the versions: 

    /opt/gitlab/bin/pg_dump --version
/opt/gitlab/bin/psql --version

    They should now be the same as your non-packaged external PostgreSQL.

After this is done, ensure that the backup and restore tasks are using the correct executables by running both the backup and restore commands.

Upgrade a non-packaged PostgreSQL database

You can upgrade the external database after stopping all the processes that are connected to the database (Puma, Sidekiq): 

sudo gitlab-ctl stop puma
sudo gitlab-ctl stop sidekiq

Before proceeding with the upgrade, note the following:

  • Check compatibility between GitLab releases and PostgreSQL versions:
    • Read about which GitLab versions introduced a requirement for a minimum PostgreSQL version.
    • Read about significant changes to the PostgreSQL versions which shipped with the Linux package: The Linux package is tested for compatibility with the major releases of PostgreSQL that it ships with.
  • When using GitLab backup or restore, you must keep the same version of GitLab. If you plan to upgrade to a later GitLab version as well, upgrade PostgreSQL first.
  • The backup and restore commands can be used to back up and restore the database to a later version of PostgreSQL.
  • If a PostgreSQL version is specified with postgresql['version'] that doesn’t ship with that Linux package release, the default version in the compatibility table determines which client binaries (such as the PostgreSQL backup/restore binaries) are active.

The following example demonstrates upgrading from a database host running PostgreSQL 13 to another database host running PostgreSQL 14 and incurs downtime:

  1. Spin up a new PostgreSQL 14 database server that’s set up according to the database requirements.

  2. Ensure that the compatible versions of pg_dump and pg_restore are being used on the GitLab Rails instance. To amend GitLab configuration, edit /etc/gitlab/gitlab.rb and specify the value of postgresql['version']: 

    postgresql['version'] = 14

  3. Reconfigure GitLab: 

    sudo gitlab-ctl reconfigure

  4. Stop GitLab (note that this step causes downtime): 

    sudo gitlab-ctl stop
caution
The backup command requires additional parameters when your installation is using PgBouncer.

  1. Run the backup Rake task using the SKIP options to back up only the database. Make a note of the backup file name; you’ll use it later to restore. 

    sudo gitlab-backup create SKIP=repositories,uploads,builds,artifacts,lfs,pages,registry

  2. Shutdown the PostgreSQL 13 database host.

  3. Edit /etc/gitlab/gitlab.rb and update the gitlab_rails['db_host'] setting to point to the PostgreSQL database 14 host.

  4. Reconfigure GitLab: 

    sudo gitlab-ctl reconfigure
    caution
    The backup command requires additional parameters when your installation is using PgBouncer.

  5. Restore the database using the database backup file created earlier, and be sure to answer no when asked “This task will now rebuild the authorized_keys file”: 

    # Use the backup timestamp https://docs.gitlab.com/ee/administration/backup_restore/backup_gitlab.html#backup-timestamp
sudo gitlab-backup restore BACKUP=<backup-timestamp>

  6. Start GitLab: 

    sudo gitlab-ctl start

  7. After upgrading PostgreSQL to a new major release, recreate the table statistics to ensure efficient query plans are picked and to reduce database server CPU load.

    If the upgrade was “in-place” using pg_upgrade, run the following query on the PostgreSQL database console: 

    SET statement_timeout = 0; ANALYZE VERBOSE;

    If the upgrade used pg_dump and pg_restore, run the following query on the PostgreSQL database console: 

    SET statement_timeout = 0; VACUUM VERBOSE ANALYZE;

Seed the database (fresh installs only)

caution
This is a destructive command; do not run it on an existing database.

The Linux package installation does not seed your external database. Run the following command to import the schema and create the first administration user: 

# Remove 'sudo' if you are the 'git' user
sudo gitlab-rake gitlab:setup

If you want to specify a password for the default root user, specify the initial_root_password setting in /etc/gitlab/gitlab.rb before running the gitlab:setup command above: 

gitlab_rails['initial_root_password'] = 'nonstandardpassword'

If you want to specify the initial registration token for shared GitLab Runners, specify the initial_shared_runners_registration_token setting in /etc/gitlab/gitlab.rb before running the gitlab:setup command: 

gitlab_rails['initial_shared_runners_registration_token'] = 'token'

Pin the packaged PostgreSQL version (fresh installs only)

The Linux package ships with different PostgreSQL versions and initializes the default version if not specified otherwise.

To initialize PostgreSQL with a non-default version, you can set postgresql['version'] to the major version one of the packaged PostgreSQL versions prior to the initial reconfigure. For example, in GitLab 15.0 you can use postgresql['version'] = 12 to use PostgreSQL 12 instead of the default of PostgreSQL 13.

caution
Setting postgresql['version'] while using the PostgreSQL packaged with the Linux package after the initial reconfigure will throw errors about the data directory being initialized on a different version of PostgreSQL. If this is encountered, see Revert packaged PostgreSQL server to the previous version.

If you are doing a fresh install on an environment that previously had GitLab installed on it and you are using a pinned PostgreSQL version, first make sure that any folders that relate to PostgreSQL are deleted and that there are no PostgreSQL processes running on the instance.

Provide sensitive data configuration to GitLab Rails without plain text storage

For more information, see the example in configuration documentation.

Application Settings for the Database

Disabling automatic database migration

If you have multiple GitLab servers sharing a database, you will want to limit the number of nodes that are performing the migration steps during reconfiguration.

Edit /etc/gitlab/gitlab.rb to append: 

# Enable or disable automatic database migrations
# on all hosts except the designated deploy node
gitlab_rails['auto_migrate'] = false

/etc/gitlab/gitlab.rb should have file permissions 0600 because it contains plain-text passwords.

The next time hosts carrying the above configuration are reconfigured, the migration steps are not performed.

To avoid schema-related post-upgrade errors, the host marked as the deploy node must have gitlab_rails['auto_migrate'] = true during upgrades.

Setting client statement_timeout

The amount of time that Rails will wait for a database transaction to complete before timing out can now be adjusted with the gitlab_rails['db_statement_timeout'] setting. By default, this setting is not used.

Edit /etc/gitlab/gitlab.rb: 

gitlab_rails['db_statement_timeout'] = 45000

In this case, the client statement_timeout is set to 45 seconds. The value is specified in milliseconds.

Setting connection timeout

The amount of time that Rails will wait for a PostgreSQL connection attempt to succeed before timing out can be adjusted with the gitlab_rails['db_connect_timeout'] setting. By default, this setting is not used:

  1. Edit /etc/gitlab/gitlab.rb: 

    gitlab_rails['db_connect_timeout'] = 5

  2. Reconfigure GitLab: 

    sudo gitlab-ctl reconfigure

In this case, the client connect_timeout is set to 5 seconds. The value is specified in seconds. A minimum value of 2 seconds applies. Setting this to <= 0 or not specifying the setting at all disables the timeout.

Setting TCP controls

The Rails PostgreSQL adapter provides a series of TCP connection controls that may be tuned to improve performance. Consult the PostgreSQL upstream documentation for more information about each parameter.

The Linux package sets no defaults for these values and instead uses the defaults provided by the PostgreSQL adapter. Override them in gitlab.rb using the parameters noted in the table below and then run gitlab-ctl reconfigure.

PostgreSQL parameter gitlab.rb parameter
keepalives gitlab_rails['db_keepalives']
keepalives_idle gitlab_rails['db_keepalives_idle']
keepalives_interval gitlab_rails['db_keepalives_interval']
keepalives_count gitlab_rails['db_keepalives_count']
tcp_user_timeout gitlab_rails['db_tcp_user_timeout']

Automatic database reindexing

caution
This is an experimental feature that isn’t enabled by default.

Recreates database indexes in the background (called “reindexing”). This can be used to remove bloated space that has accumulated in indexes and helps to maintain healthy and efficient indexes.

The reindexing task can be started regularly through a cronjob. To configure the cronjob, gitlab_rails['database_reindexing']['enable'] should be set to true.

In a multi-node environment, this feature should only be enabled on an application host. The reindexing process cannot go through PgBouncer, it has to have a direct database connection.

By default, this starts the cronjob every hour during weekends (likely a low-traffic time) only.

You can change the schedule by refining the following settings:

  1. Edit /etc/gitlab/gitlab.rb: 

    gitlab_rails['database_reindexing']['hour'] = '*'
gitlab_rails['database_reindexing']['minute'] = 0
gitlab_rails['database_reindexing']['month'] = '*'
gitlab_rails['database_reindexing']['day_of_month'] = '*'
gitlab_rails['database_reindexing']['day_of_week'] = '0,6'

  2. Reconfigure GitLab: 

    sudo gitlab-ctl reconfigure

Packaged PostgreSQL deployed in an HA/Geo Cluster

Upgrading a GitLab HA cluster

To upgrade the PostgreSQL version in a Patroni cluster see Upgrading PostgreSQL major version in a Patroni cluster.

Upgrading a GitLab HA Repmgr cluster

note
If you are upgrading to PostgreSQL 12, you need to switch from Repmgr to Patroni first see Switching from Repmgr to Patroni.

These instructions are provided for upgrading an older GitLab cluster to PostgreSQL 11 when using Repmgr.

If PostgreSQL is configured for high availability, pg-upgrade should be run on all the nodes running PostgreSQL. Other nodes can be skipped but must be running the same GitLab version as the database nodes.

Follow the steps below to upgrade the database nodes:

  1. Secondary nodes must be upgraded before the primary node.

    1. On the secondary nodes, edit /etc/gitlab/gitlab.rb to include the following: 

      # Replace X with the number of DB nodes + 1
postgresql['max_replication_slots'] = X
    2. Run gitlab-ctl reconfigure to update the configuration.
    3. Run sudo gitlab-ctl restart postgresql to get PostgreSQL restarted with the new configuration.
    4. On running pg-upgrade on a PostgreSQL secondary node, the node will be removed from the cluster.
    5. Once all the secondary nodes are upgraded using pg-upgrade, the user will be left with a single-node cluster that has only the primary node.
    6. pg-upgrade, on secondary nodes will not update the existing data to match the new version, as that data will be replaced by the data from the primary node. It will however move the existing data to a backup location.
  2. Once all secondary nodes are upgraded, run pg-upgrade on the primary node.

    1. On the primary node, edit /etc/gitlab/gitlab.rb to include the following: 

      # Replace X with the number of DB nodes + 1
postgresql['max_replication_slots'] = X
    2. Run gitlab-ctl reconfigure to update the configuration.
    3. Run sudo gitlab-ctl restart postgresql to get PostgreSQL restarted with the new configuration.
    4. On a primary node, pg-upgrade will update the existing data to match the new PostgreSQL version.

  3. Recreate the secondary nodes by running the following command on each of them 

    gitlab-ctl repmgr standby setup MASTER_NODE_NAME

  4. Check if the repmgr cluster is back to the original state 

    gitlab-ctl repmgr cluster show

Troubleshooting upgrades in an HA cluster

If at some point, the bundled PostgreSQL had been running on a node before upgrading to an HA setup, the old data directory may remain. This will cause gitlab-ctl reconfigure to downgrade the version of the PostgreSQL utilities it uses on that node. Move (or remove) the directory to prevent this:

  • mv /var/opt/gitlab/postgresql/data/ /var/opt/gitlab/postgresql/data.$(date +%s)

If you encounter the following error when recreating the secondary nodes with gitlab-ctl repmgr standby setup MASTER_NODE_NAME, ensure that you have postgresql['max_replication_slots'] = X (where X is the number of DB nodes + 1), is included in /etc/gitlab/gitlab.rb: 

pg_basebackup: could not create temporary replication slot "pg_basebackup_12345": ERROR:  all replication slots are in use
HINT:  Free one or increase max_replication_slots.

Upgrading a Geo instance

Since Geo depends on PostgreSQL streaming replication by default, there are additional considerations when upgrading GitLab and/or when upgrading PostgreSQL is described below.

Caveats when upgrading PostgreSQL with Geo

caution
When using Geo, upgrading PostgreSQL requires downtime on all secondaries.

When using Geo, upgrading PostgreSQL requires downtime on all secondaries because it requires re-initializing PostgreSQL replication to Geo secondaries. This is due to the way PostgreSQL streaming replication works. Re-initializing replication copies all data from the primary again, so it can take a long time depending mostly on the size of the database and available bandwidth. For example, at a transfer speed of 30 Mbps, and a database size of 100 GB, resynchronization could take approximately 8 hours. See PostgreSQL documentation for more.

How to upgrade PostgreSQL when using Geo

To upgrade PostgreSQL, you will need the name of the replication slot, and the replication user’s password.

  1. Find the name of the existing replication slot on the Geo primary’s database node, run: 

    sudo gitlab-psql -qt -c 'select slot_name from pg_replication_slots'

    If you can’t find your slot_name here, or there is no output returned, your Geo secondaries may not be healthy. In that case, make sure the secondaries are healthy and replication is working.

    Even if the query is empty, you can try to reinitialize the secondary database with the slot_name found on the Geo sites admin area.

  2. Gather the replication user’s password. It was set while setting up Geo in Step 1. Configure the primary site.

  3. Manually upgrade PostgreSQL on the Geo primary. Run on the Geo primary’s database node: 

    sudo gitlab-ctl pg-upgrade

    Wait for the primary database to finish upgrading before beginning the following step, so the secondary can remain ready as a backup. Afterward, you can upgrade the tracking database in parallel with the secondary database.

  4. Manually upgrade PostgreSQL on the Geo secondaries. Run on the Geo secondary database and also on the tracking database: 

    sudo gitlab-ctl pg-upgrade

  5. Restart the database replication on the Geo secondary database using the command: 

    sudo gitlab-ctl replicate-geo-database --slot-name=SECONDARY_SLOT_NAME --host=PRIMARY_HOST_NAME --sslmode=verify-ca

    You will be prompted for the replication user’s password of the primary. Replace SECONDARY_SLOT_NAME with the slot name retrieved from the first step above.

  6. Reconfigure GitLab on the Geo secondary database to update the pg_hba.conf file. This is needed because replicate-geo-database replicates the primary’s file to the secondary.

  7. Restart puma, sidekiq, and geo-logcursor. 

    sudo gitlab-ctl hup puma
sudo gitlab-ctl restart sidekiq
sudo gitlab-ctl restart geo-logcursor

  8. Navigate to https://your_primary_server/admin/geo/nodes and ensure that all nodes are healthy.

Connecting to the PostgreSQL database

If you need to connect to the PostgreSQL database, you can connect as the application user: 

sudo gitlab-rails dbconsole --database main

Troubleshooting

Set default_transaction_isolation into read committed

If you see errors similar to the following in your production/sidekiq log: 

ActiveRecord::StatementInvalid PG::TRSerializationFailure: ERROR:  could not serialize access due to concurrent update

Chances are your database’s default_transaction_isolation configuration is not in line with the GitLab application requirement. You can check this configuration by connecting to your PostgreSQL database and run SHOW default_transaction_isolation;. GitLab application expects read committed to be configured.

This default_transaction_isolation configuration is set in your postgresql.conf file. You will need to restart/reload the database once you changed the configuration. This configuration comes by default in the packaged PostgreSQL server included with the Linux package.

Could not load library plpgsql.so

You might see errors similar to the following while running Database migrations or in the PostgreSQL/Patroni logs: 

ERROR:  could not load library "/opt/gitlab/embedded/postgresql/12/lib/plpgsql.so": /opt/gitlab/embedded/postgresql/12/lib/plpgsql.so: undefined symbol: EnsurePortalSnapshotExists

This error is caused due to not restarting PostgreSQL after the underlying version changed. To fix this error:

  1. Run one of the following commands: 

    # For PostgreSQL
sudo gitlab-ctl restart postgresql

# For Patroni
sudo gitlab-ctl restart patroni

# For Geo PostgreSQL
sudo gitlab-ctl restart geo-postgresql

  2. Reconfigure GitLab: 

    sudo gitlab-ctl reconfigure

Database CPU load very high

If the database CPU load is very high, it could be caused by the auto cancel redundant pipelines setting. For more details, see issue 435250.

To work around this issue:

  • You can allocate more CPU resources to the database server.
  • If Sidekiq is overloaded, you might need to add more Sidekiq processes for the ci_cancel_redundant_pipelines queue if your projects have a very large number of pipelines.
  • You can enable the disable_cancel_redundant_pipelines_service feature flag to disable this setting instance-wide and see if the CPU load goes down. This disables the feature for all projects, and can lead to increased resource use by pipelines that are no longer being cancelled automatically.