Monitoring GitLab with Prometheus

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

Prometheus is a powerful time-series monitoring service, providing a flexible platform for monitoring GitLab and other software products.

GitLab provides out-of-the-box monitoring with Prometheus, providing access to high quality time-series monitoring of GitLab services.

Prometheus and the various exporters listed in this page are bundled in Linux packages. Check each exporter’s documentation for the timeline they got added. For self-compiled installations, you must install them yourself. Over subsequent releases additional GitLab metrics are captured.

Prometheus services are on by default.

Prometheus and its exporters don’t authenticate users, and are available to anyone who can access them.

How Prometheus works

Prometheus works by periodically connecting to data sources and collecting their performance metrics through the various exporters. To view and work with the monitoring data, you can either connect directly to Prometheus or use a dashboard tool like Grafana.

Configuring Prometheus

For self-compiled installations, you must install and configure it yourself.

Prometheus and its exporters are on by default. Prometheus runs as the gitlab-prometheus user and listen on http://localhost:9090. By default, Prometheus is only accessible from the GitLab server itself. Each exporter is automatically set up as a monitoring target for Prometheus, unless individually disabled.

To disable Prometheus and all of its exporters, as well as any added in the future:

  1. Edit /etc/gitlab/gitlab.rb
  2. Add or find and uncomment the following lines, making sure they are set to false:

    prometheus_monitoring['enable'] = false
    sidekiq['metrics_enabled'] = false
    
    # Already set to `false` by default, but you can explicitly disable it to be sure
    puma['exporter_enabled'] = false
    
  3. Save the file and reconfigure GitLab for the changes to take effect.

Changing the port and address Prometheus listens on

caution
Although possible, it’s not recommended to change the port Prometheus listens on, as this might affect or conflict with other services running on the GitLab server. Proceed at your own risk.

To access Prometheus from outside the GitLab server, change the address/port that Prometheus listens on:

  1. Edit /etc/gitlab/gitlab.rb
  2. Add or find and uncomment the following line:

    prometheus['listen_address'] = 'localhost:9090'
    

    Replace localhost:9090 with the address or port you want Prometheus to listen on. If you would like to allow access to Prometheus to hosts other than localhost, leave out the host, or use 0.0.0.0 to allow public access:

    prometheus['listen_address'] = ':9090'
    # or
    prometheus['listen_address'] = '0.0.0.0:9090'
    
  3. Save the file and reconfigure GitLab for the changes to take effect

Adding custom scrape configurations

You can configure additional scrape targets for the Linux package-bundled Prometheus by editing prometheus['scrape_configs'] in /etc/gitlab/gitlab.rb using the Prometheus scrape target configuration syntax.

Here is an example configuration to scrape http://1.1.1.1:8060/probe?param_a=test&param_b=additional_test:

prometheus['scrape_configs'] = [
  {
    'job_name': 'custom-scrape',
    'metrics_path': '/probe',
    'params' => {
      'param_a' => ['test'],
      'param_b' => ['additional_test'],
    },
    'static_configs' => [
      'targets' => ['1.1.1.1:8060'],
    ],
  },
]

Standalone Prometheus using the Linux package

The Linux package can be used to configure a standalone Monitoring node running Prometheus and Grafana. A standalone Monitoring node is recommended for GitLab deployments with multiple nodes.

The steps below are the minimum necessary to configure a Monitoring node running Prometheus and Grafana with the Linux package:

  1. SSH into the Monitoring node.
  2. Install the Linux package you want using steps 1 and 2 from the GitLab downloads page, but do not follow the remaining steps.
  3. Make sure to collect the IP addresses or DNS records of the Consul server nodes, for the next step.
  4. Edit /etc/gitlab/gitlab.rb and add the contents:

    roles ['monitoring_role']
    
    external_url 'http://gitlab.example.com'
    
    # Prometheus
    prometheus['listen_address'] = '0.0.0.0:9090'
    prometheus['monitor_kubernetes'] = false
    
    # Grafana
    grafana['enable'] = true
    grafana['admin_password'] = 'toomanysecrets'
    grafana['disable_login_form'] = false
    
    # Enable service discovery for Prometheus
    consul['enable'] = true
    consul['monitoring_service_discovery'] = true
    consul['configuration'] = {
       retry_join: %w(10.0.0.1 10.0.0.2 10.0.0.3), # The addresses can be IPs or FQDNs
    }
    
    # Nginx - For Grafana access
    nginx['enable'] = true
    
  5. Run sudo gitlab-ctl reconfigure to compile the configuration.

The next step is to tell all the other nodes where the monitoring node is:

  1. Edit /etc/gitlab/gitlab.rb, and add, or find and uncomment the following line:

    # can be FQDN or IP
    gitlab_rails['prometheus_address'] = '10.0.0.1:9090'
    

    Where 10.0.0.1:9090 is the IP address and port of the Prometheus node.

  2. Save the file and reconfigure GitLab for the changes to take effect.

After monitoring using Service Discovery is enabled with consul['monitoring_service_discovery'] = true, ensure that prometheus['scrape_configs'] is not set in /etc/gitlab/gitlab.rb. Setting both consul['monitoring_service_discovery'] = true and prometheus['scrape_configs'] in /etc/gitlab/gitlab.rb results in errors.

Using an external Prometheus server

caution
Prometheus and most exporters don’t support authentication. We don’t recommend exposing them outside the local network.

A few configuration changes are required to allow GitLab to be monitored by an external Prometheus server.

To use an external Prometheus server:

  1. Edit /etc/gitlab/gitlab.rb.
  2. Disable the bundled Prometheus:

    prometheus['enable'] = false
    
  3. Set each bundled service’s exporter to listen on a network address, for example:

    node_exporter['listen_address'] = '0.0.0.0:9100'
    gitlab_workhorse['prometheus_listen_addr'] = "0.0.0.0:9229"
    
    # Rails nodes
    gitlab_exporter['listen_address'] = '0.0.0.0'
    gitlab_exporter['listen_port'] = '9168'
    registry['debug_addr'] = '0.0.0.0:5001'
    
    # Sidekiq nodes
    sidekiq['listen_address'] = '0.0.0.0'
    
    # Redis nodes
    redis_exporter['listen_address'] = '0.0.0.0:9121'
    
    # PostgreSQL nodes
    postgres_exporter['listen_address'] = '0.0.0.0:9187'
    
    # Gitaly nodes
    gitaly['configuration'] = {
       # ...
       prometheus_listen_addr: '0.0.0.0:9236',
    }
    
    # Pgbouncer nodes
    pgbouncer_exporter['listen_address'] = '0.0.0.0:9188'
    
  4. Install and set up a dedicated Prometheus instance, if necessary, using the official installation instructions.

  5. On all GitLab Rails (Puma, Sidekiq) servers, set the Prometheus server IP address and listen port. For example:

    gitlab_rails['prometheus_address'] = '192.168.0.1:9090'
    
  6. To scrape NGINX metrics, you must also configure NGINX to allow the Prometheus server IP. For example:

    nginx['status']['options'] = {
          "server_tokens" => "off",
          "access_log" => "off",
          "allow" => "192.168.0.1",
          "deny" => "all",
    }
    

    You can also specify more than one IP address if you have multiple Prometheus servers:

    nginx['status']['options'] = {
          "server_tokens" => "off",
          "access_log" => "off",
          "allow" => ["192.168.0.1", "192.168.0.2"],
          "deny" => "all",
    }
    
  7. To allow the Prometheus server to fetch from the GitLab metrics endpoint, add the Prometheus server IP address to the monitoring IP allowlist:

    gitlab_rails['monitoring_whitelist'] = ['127.0.0.0/8', '192.168.0.1']
    
  8. As we are setting each bundled service’s exporter to listen on a network address, update the firewall on the instance to only allow traffic from your Prometheus IP for the exporters enabled. A full reference list of exporter services and their respective ports can be found here.
  9. Reconfigure GitLab to apply the changes.
  10. Edit the Prometheus server’s configuration file.
  11. Add each node’s exporters to the Prometheus server’s scrape target configuration. For example, a sample snippet using static_configs:

    scrape_configs:
      - job_name: nginx
        static_configs:
          - targets:
            - 1.1.1.1:8060
      - job_name: redis
        static_configs:
          - targets:
            - 1.1.1.1:9121
      - job_name: postgres
        static_configs:
          - targets:
            - 1.1.1.1:9187
      - job_name: node
        static_configs:
          - targets:
            - 1.1.1.1:9100
      - job_name: gitlab-workhorse
        static_configs:
          - targets:
            - 1.1.1.1:9229
      - job_name: gitlab-rails
        metrics_path: "/-/metrics"
        scheme: https
        static_configs:
          - targets:
            - 1.1.1.1
      - job_name: gitlab-sidekiq
        static_configs:
          - targets:
            - 1.1.1.1:8082
      - job_name: gitlab_exporter_database
        metrics_path: "/database"
        static_configs:
          - targets:
            - 1.1.1.1:9168
      - job_name: gitlab_exporter_sidekiq
        metrics_path: "/sidekiq"
        static_configs:
          - targets:
            - 1.1.1.1:9168
      - job_name: gitaly
        static_configs:
          - targets:
            - 1.1.1.1:9236
      - job_name: registry
        static_configs:
          - targets:
            - 1.1.1.1:5001
    
    caution
    The gitlab-rails job in the snippet assumes that GitLab is reachable through HTTPS. If your deployment doesn’t use HTTPS, the job configuration is adapted to use the http scheme and port 80.
  12. Reload the Prometheus server.

Configure the storage retention size

Prometheus has several custom flags to configure local storage:

  • storage.tsdb.retention.time: when to remove old data. Defaults to 15d. Overrides storage.tsdb.retention if this flag is set to anything other than the default.
  • storage.tsdb.retention.size: (experimental) the maximum number of bytes of storage blocks to retain. The oldest data is removed first. Defaults to 0 (disabled). This flag is experimental and may change in future releases. Units supported: B, KB, MB, GB, TB, PB, EB. For example, 512MB.

To configure the storage retention size:

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

    prometheus['flags'] = {
      'storage.tsdb.path' => "/var/opt/gitlab/prometheus/data",
      'storage.tsdb.retention.time' => "7d",
      'storage.tsdb.retention.size' => "2GB",
      'config.file' => "/var/opt/gitlab/prometheus/prometheus.yml"
    }
    
  2. Reconfigure GitLab:

    sudo gitlab-ctl reconfigure
    

Viewing performance metrics

You can visit http://localhost:9090 for the dashboard that Prometheus offers by default.

If SSL has been enabled on your GitLab instance, you may not be able to access Prometheus on the same browser as GitLab if using the same FQDN due to HSTS. A test project exists to provide access via GitLab, but in the interim there are some workarounds: using a separate FQDN, using server IP, using a separate browser for Prometheus, resetting HSTS, or having NGINX proxy it.

The performance data collected by Prometheus can be viewed directly in the Prometheus console, or through a compatible dashboard tool. The Prometheus interface provides a flexible query language to work with the collected data where you can visualize the output. For a more fully featured dashboard, Grafana can be used and has official support for Prometheus.

Sample Prometheus queries:

  • % Memory available: ((node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes) or ((node_memory_MemFree_bytes + node_memory_Buffers_bytes + node_memory_Cached_bytes) / node_memory_MemTotal_bytes)) * 100
  • % CPU utilization: 1 - avg without (mode,cpu) (rate(node_cpu_seconds_total{mode="idle"}[5m]))
  • Data transmitted: rate(node_network_transmit_bytes_total{device!="lo"}[5m])
  • Data received: rate(node_network_receive_bytes_total{device!="lo"}[5m])

Prometheus as a Grafana data source

Grafana allows you to import Prometheus performance metrics as a data source, and render the metrics as graphs and dashboards, which is helpful with visualization.

To add a Prometheus dashboard for a single server GitLab setup:

  1. Create a new data source in Grafana.
  2. Name your data source (such as GitLab).
  3. Select Prometheus in the type dropdown list.
  4. Add your Prometheus listen address as the URL, and set access to Browser.
  5. Set the HTTP method to GET.
  6. Save and test your configuration to verify that it works.

GitLab metrics

GitLab monitors its own internal service metrics, and makes them available at the /-/metrics endpoint. Unlike other exporters, this endpoint requires authentication as it’s available on the same URL and port as user traffic.

Read more about the GitLab Metrics.

Bundled software metrics

Many of the GitLab dependencies bundled in the Linux package are preconfigured to export Prometheus metrics.

Node exporter

The node exporter allows you to measure various machine resources, such as memory, disk, and CPU utilization.

Read more about the node exporter.

Web exporter

The web exporter is a dedicated metrics server that allows splitting end-user and Prometheus traffic into two separate applications to improve performance and availability.

Read more about the web exporter.

Redis exporter

The Redis exporter allows you to measure various Redis metrics.

Read more about the Redis exporter.

PostgreSQL exporter

The PostgreSQL exporter allows you to measure various PostgreSQL metrics.

Read more about the PostgreSQL exporter.

PgBouncer exporter

The PgBouncer exporter allows you to measure various PgBouncer metrics.

Read more about the PgBouncer exporter.

Registry exporter

The Registry exporter allows you to measure various Registry metrics.

Read more about the Registry exporter.

GitLab exporter

The GitLab exporter allows you to measure various GitLab metrics, pulled from Redis and the database.

Read more about the GitLab exporter.

Troubleshooting

/var/opt/gitlab/prometheus consumes too much disk space

If you are not using Prometheus monitoring:

  1. Disable Prometheus.
  2. Delete the data under /var/opt/gitlab/prometheus.

If you are using Prometheus monitoring:

  1. Stop Prometheus (deleting data while it’s running can cause data corruption):

    gitlab-ctl stop prometheus
    
  2. Delete the data under /var/opt/gitlab/prometheus/data.
  3. Start the service again:

    gitlab-ctl start prometheus
    
  4. Verify the service is up and running:

    gitlab-ctl status prometheus
    
  5. Optional. Configure the storage retention size.

Monitoring node not receiving data

If the monitoring node is not receiving any data, check that the exporters are capturing data:

curl "http[s]://localhost:<EXPORTER LISTENING PORT>/metrics"

or

curl "http[s]://localhost:<EXPORTER LISTENING PORT>/-/metrics"