Working with the bundled PgBouncer service

PgBouncer is used to seamlessly migrate database connections between servers in a failover scenario. Additionally, it can be used in a non-fault-tolerant setup to pool connections, speeding up response time while reducing resource usage.

GitLab Premium includes a bundled version of PgBouncer that can be managed through /etc/gitlab/gitlab.rb.

PgBouncer as part of a fault-tolerant GitLab installation

This content has been moved to a new location.

PgBouncer as part of a non-fault-tolerant GitLab installation

  1. Generate PGBOUNCER_USER_PASSWORD_HASH with the command gitlab-ctl pg-password-md5 pgbouncer

  2. Generate SQL_USER_PASSWORD_HASH with the command gitlab-ctl pg-password-md5 gitlab. Enter the plaintext SQL_USER_PASSWORD later.

  3. On your database node, ensure the following is set in your /etc/gitlab/gitlab.rb 

    postgresql['pgbouncer_user_password'] = 'PGBOUNCER_USER_PASSWORD_HASH'
postgresql['sql_user_password'] = 'SQL_USER_PASSWORD_HASH'
postgresql['listen_address'] = 'XX.XX.XX.Y' # Where XX.XX.XX.Y is the ip address on the node postgresql should listen on
postgresql['md5_auth_cidr_addresses'] = %w(AA.AA.AA.B/32) # Where AA.AA.AA.B is the IP address of the pgbouncer node

  4. Run gitlab-ctl reconfigure

    note
    If the database was already running, it needs to be restarted after reconfigure by running gitlab-ctl restart postgresql.

  5. On the node you are running PgBouncer on, make sure the following is set in /etc/gitlab/gitlab.rb 

    pgbouncer['enable'] = true
pgbouncer['databases'] = {
  gitlabhq_production: {
    host: 'DATABASE_HOST',
    user: 'pgbouncer',
    password: 'PGBOUNCER_USER_PASSWORD_HASH'
  }
}

    You can pass additional configuration parameters per database, for example: 

    pgbouncer['databases'] = {
  gitlabhq_production: {
     ...
     pool_mode: 'transaction'
  }
}

    Use these parameters with caution. For the complete list of parameters refer to the PgBouncer documentation.

  6. Run gitlab-ctl reconfigure

  7. On the node running Puma, make sure the following is set in /etc/gitlab/gitlab.rb 

    gitlab_rails['db_host'] = 'PGBOUNCER_HOST'
gitlab_rails['db_port'] = '6432'
gitlab_rails['db_password'] = 'SQL_USER_PASSWORD'

  8. Run gitlab-ctl reconfigure

  9. At this point, your instance should connect to the database through PgBouncer. If you are having issues, see the Troubleshooting section

Backups

Do not backup or restore GitLab through a PgBouncer connection: it causes a GitLab outage.

Read more about this and how to reconfigure backups.

Enable Monitoring

Introduced in GitLab 12.0.

If you enable Monitoring, it must be enabled on all PgBouncer servers.

  1. Create/edit /etc/gitlab/gitlab.rb and add the following configuration: 

    # Enable service discovery for Prometheus
consul['enable'] = true
consul['monitoring_service_discovery'] =  true

# Replace placeholders
# Y.Y.Y.Y consul1.gitlab.example.com Z.Z.Z.Z
# with the addresses of the Consul server nodes
consul['configuration'] = {
   retry_join: %w(Y.Y.Y.Y consul1.gitlab.example.com Z.Z.Z.Z),
}

# Set the network addresses that the exporters will listen on
node_exporter['listen_address'] = '0.0.0.0:9100'
pgbouncer_exporter['listen_address'] = '0.0.0.0:9188'

  2. Run sudo gitlab-ctl reconfigure to compile the configuration.

Administrative console

As part of Omnibus GitLab, a command is provided to automatically connect to the PgBouncer administrative console. See the PgBouncer documentation for detailed instructions on how to interact with the console.

To start a session run the following and provide the password for the pgbouncer user: 

sudo gitlab-ctl pgb-console

To get some basic information about the instance: 

pgbouncer=# show databases; show clients; show servers;
        name         |   host    | port |      database       | force_user | pool_size | reserve_pool | pool_mode | max_connections | current_connections
---------------------+-----------+------+---------------------+------------+-----------+--------------+-----------+-----------------+---------------------
 gitlabhq_production | 127.0.0.1 | 5432 | gitlabhq_production |            |       100 |            5 |           |               0 |                   1
 pgbouncer           |           | 6432 | pgbouncer           | pgbouncer  |         2 |            0 | statement |               0 |                   0
(2 rows)

 type |   user    |      database       | state  |   addr    | port  | local_addr | local_port |    connect_time     |    request_time     |    ptr    | link
| remote_pid | tls
------+-----------+---------------------+--------+-----------+-------+------------+------------+---------------------+---------------------+-----------+------
+------------+-----
 C    | gitlab    | gitlabhq_production | active | 127.0.0.1 | 44590 | 127.0.0.1  |       6432 | 2018-04-24 22:13:10 | 2018-04-24 22:17:10 | 0x12444c0 |
|          0 |
 C    | gitlab    | gitlabhq_production | active | 127.0.0.1 | 44592 | 127.0.0.1  |       6432 | 2018-04-24 22:13:10 | 2018-04-24 22:17:10 | 0x12447c0 |
|          0 |
 C    | gitlab    | gitlabhq_production | active | 127.0.0.1 | 44594 | 127.0.0.1  |       6432 | 2018-04-24 22:13:10 | 2018-04-24 22:17:10 | 0x1244940 |
|          0 |
 C    | gitlab    | gitlabhq_production | active | 127.0.0.1 | 44706 | 127.0.0.1  |       6432 | 2018-04-24 22:14:22 | 2018-04-24 22:16:31 | 0x1244ac0 |
|          0 |
 C    | gitlab    | gitlabhq_production | active | 127.0.0.1 | 44708 | 127.0.0.1  |       6432 | 2018-04-24 22:14:22 | 2018-04-24 22:15:15 | 0x1244c40 |
|          0 |
 C    | gitlab    | gitlabhq_production | active | 127.0.0.1 | 44794 | 127.0.0.1  |       6432 | 2018-04-24 22:15:15 | 2018-04-24 22:15:15 | 0x1244dc0 |
|          0 |
 C    | gitlab    | gitlabhq_production | active | 127.0.0.1 | 44798 | 127.0.0.1  |       6432 | 2018-04-24 22:15:15 | 2018-04-24 22:16:31 | 0x1244f40 |
|          0 |
 C    | pgbouncer | pgbouncer           | active | 127.0.0.1 | 44660 | 127.0.0.1  |       6432 | 2018-04-24 22:13:51 | 2018-04-24 22:17:12 | 0x1244640 |
|          0 |
(8 rows)

 type |  user  |      database       | state |   addr    | port | local_addr | local_port |    connect_time     |    request_time     |    ptr    | link | rem
ote_pid | tls
------+--------+---------------------+-------+-----------+------+------------+------------+---------------------+---------------------+-----------+------+----
--------+-----
 S    | gitlab | gitlabhq_production | idle  | 127.0.0.1 | 5432 | 127.0.0.1  |      35646 | 2018-04-24 22:15:15 | 2018-04-24 22:17:10 | 0x124dca0 |      |
  19980 |
(1 row)

Procedure for bypassing PgBouncer

Some database changes have to be done directly, and not through PgBouncer.

Read more about the affected tasks: database restores and GitLab upgrades.

  1. To find the primary node, run the following on a database node: 

    sudo gitlab-ctl patroni members

  2. Edit /etc/gitlab/gitlab.rb on the application node you’re performing the task on, and update gitlab_rails['db_host'] and gitlab_rails['db_port'] with the database primary’s host and port.

  3. Run reconfigure: 

    sudo gitlab-ctl reconfigure

Once you’ve performed the tasks or procedure, switch back to using PgBouncer:

  1. Change back /etc/gitlab/gitlab.rb to point to PgBouncer.

  2. Run reconfigure: 

    sudo gitlab-ctl reconfigure

Fine tuning

PgBouncer’s default settings suit the majority of installations. In specific cases you may want to change the performance-specific and resource-specific variables to either increase possible throughput or to limit resource utilization that could cause memory exhaustion on the database.

You can find the parameters and respective documentation on the official PgBouncer documentation. Listed below are the most relevant ones and their defaults on an Omnibus GitLab installation:

  • pgbouncer['max_client_conn'] (default: 2048, depends on server file descriptor limits) This is the “frontend” pool in PgBouncer: connections from Rails to PgBouncer.
  • pgbouncer['default_pool_size'] (default: 100) This is the “backend” pool in PgBouncer: connections from PgBouncer to the database.

The ideal number for default_pool_size must be enough to handle all provisioned services that need to access the database. Each of the listed services below use the following formula to define database pool size:

  • puma : max_threads + headroom (default 14)
    • max_threads is configured via: gitlab['puma']['max_threads'] (default: 4)
    • headroom can be configured via DB_POOL_HEADROOM environment variable (default to 10)
  • sidekiq : max_concurrency + 1 + headroom (default: 31)
    • max_concurrency is configured via: sidekiq['max_concurrency'] (default: 20)
    • headroom can be configured via DB_POOL_HEADROOM environment variable (default to 10)
  • geo-logcursor: 1+headroom (default: 11)
    • headroom can be configured via DB_POOL_HEADROOM environment variable (default to 10)

To calculate the default_pool_size, multiply the number of instances of puma, sidekiq and geo-logcursor by the number of connections each can consume as per listed above. The total is the suggested default_pool_size.

If you are using more than one PgBouncer with an internal Load Balancer, you may be able to divide the default_pool_size by the number of instances to guarantee an evenly distributed load between them.

The pgbouncer['max_client_conn'] is the hard limit of connections PgBouncer can accept. It’s unlikely you need to change this. If you are hitting that limit, you may want to consider adding additional PgBouncers with an internal Load Balancer.

When setting up the limits for a PgBouncer that points to the Geo Tracking Database, you can likely ignore puma from the equation, as it is only accessing that database sporadically.

Troubleshooting

In case you are experiencing any issues connecting through PgBouncer, the first place to check is always the logs: 

sudo gitlab-ctl tail pgbouncer

Additionally, you can check the output from show databases in the administrative console. In the output, you would expect to see values in the host field for the gitlabhq_production database. Additionally, current_connections should be greater than 1.

Message: LOG: invalid CIDR mask in address

See the suggested fix in Geo documentation.

Message: LOG: invalid IP mask "md5": Name or service not known

See the suggested fix in Geo documentation.