Configure Gitaly Cluster (FREE SELF)

Configure Gitaly Cluster using either:

  • The Gitaly Cluster configuration instructions available as part of reference architectures for installations for more than 2000 users.
  • The advanced configuration instructions that follow on this page.

Smaller GitLab installations may need only Gitaly itself.

Requirements for configuring a Gitaly Cluster

The minimum recommended configuration for a Gitaly Cluster requires:

  • 1 load balancer
  • 1 PostgreSQL server (PostgreSQL 11 or newer)
  • 3 Praefect nodes
  • 3 Gitaly nodes (1 primary, 2 secondary)

See the design document for implementation details.

NOTE: If not set in GitLab, feature flags are read as false from the console and Praefect uses their default value. The default value depends on the GitLab version.

Setup Instructions

If you installed GitLab using the Omnibus package (highly recommended), follow the steps below:

  1. Preparation
  2. Configuring the Praefect database
  3. Configuring the Praefect proxy/router
  4. Configuring each Gitaly node (once for each Gitaly node)
  5. Configure the load balancer
  6. Updating the GitLab server configuration
  7. Configure Grafana

Preparation

Before beginning, you should already have a working GitLab instance. Learn how to install GitLab.

Provision a PostgreSQL server (PostgreSQL 11 or newer).

Prepare all your new nodes by installing GitLab.

  • At least 1 Praefect node (minimal storage required)
  • 3 Gitaly nodes (high CPU, high memory, fast storage)
  • 1 GitLab server

You need the IP/host address for each node.

  1. LOAD_BALANCER_SERVER_ADDRESS: the IP/host address of the load balancer
  2. POSTGRESQL_SERVER_ADDRESS: the IP/host address of the PostgreSQL server
  3. PRAEFECT_HOST: the IP/host address of the Praefect server
  4. GITALY_HOST_*: the IP or host address of each Gitaly server
  5. GITLAB_HOST: the IP/host address of the GitLab server

If you are using a cloud provider, you can look up the addresses for each server through your cloud provider's management console.

If you are using Google Cloud Platform, SoftLayer, or any other vendor that provides a virtual private cloud (VPC) you can use the private addresses for each cloud instance (corresponds to "internal address" for Google Cloud Platform) for PRAEFECT_HOST, GITALY_HOST_*, and GITLAB_HOST.

Secrets

The communication between components is secured with different secrets, which are described below. Before you begin, generate a unique secret for each, and make note of it. This enables you to replace these placeholder tokens with secure tokens as you complete the setup process.

  1. GITLAB_SHELL_SECRET_TOKEN: this is used by Git hooks to make callback HTTP API requests to GitLab when accepting a Git push. This secret is shared with GitLab Shell for legacy reasons.
  2. PRAEFECT_EXTERNAL_TOKEN: repositories hosted on your Praefect cluster can only be accessed by Gitaly clients that carry this token.
  3. PRAEFECT_INTERNAL_TOKEN: this token is used for replication traffic inside your Praefect cluster. This is distinct from PRAEFECT_EXTERNAL_TOKEN because Gitaly clients must not be able to access internal nodes of the Praefect cluster directly; that could lead to data loss.
  4. PRAEFECT_SQL_PASSWORD: this password is used by Praefect to connect to PostgreSQL.

We note in the instructions below where these secrets are required.

NOTE: Omnibus GitLab installations can use gitlab-secrets.json for GITLAB_SHELL_SECRET_TOKEN.

PostgreSQL

NOTE: Do not store the GitLab application database and the Praefect database on the same PostgreSQL server if using Geo. The replication state is internal to each instance of GitLab and should not be replicated.

These instructions help set up a single PostgreSQL database, which creates a single point of failure. The following options are available:

To complete this section you need:

  • 1 Praefect node
  • 1 PostgreSQL server (PostgreSQL 11 or newer)
    • An SQL user with permissions to create databases

During this section, we configure the PostgreSQL server, from the Praefect node, using psql which is installed by Omnibus GitLab.

  1. SSH into the Praefect node and login as root:

    sudo -i
  2. Connect to the PostgreSQL server with administrative access. This is likely the postgres user. The database template1 is used because it is created by default on all PostgreSQL servers.

    /opt/gitlab/embedded/bin/psql -U postgres -d template1 -h POSTGRESQL_SERVER_ADDRESS

    Create a new user praefect to be used by Praefect. Replace PRAEFECT_SQL_PASSWORD with the strong password you generated in the preparation step.

    CREATE ROLE praefect WITH LOGIN CREATEDB PASSWORD 'PRAEFECT_SQL_PASSWORD';
  3. Reconnect to the PostgreSQL server, this time as the praefect user:

    /opt/gitlab/embedded/bin/psql -U praefect -d template1 -h POSTGRESQL_SERVER_ADDRESS

    Create a new database praefect_production. By creating the database while connected as the praefect user, we are confident they have access.

    CREATE DATABASE praefect_production WITH ENCODING=UTF8;

The database used by Praefect is now configured.

PgBouncer

To reduce PostgreSQL resource consumption, we recommend setting up and configuring PgBouncer in front of the PostgreSQL instance. To do this, set the corresponding IP or host address of the PgBouncer instance in /etc/gitlab/gitlab.rb by changing the following settings:

  • praefect['database_host'], for the address.
  • praefect['database_port'], for the port.

Because PgBouncer manages resources more efficiently, Praefect still requires a direct connection to the PostgreSQL database. It uses the LISTEN feature that is not supported by PgBouncer with pool_mode = transaction. Set praefect['database_host_no_proxy'] and praefect['database_port_no_proxy'] to a direct connection, and not a PgBouncer connection.

Save the changes to /etc/gitlab/gitlab.rb and reconfigure Praefect.

This documentation doesn't provide PgBouncer installation instructions, but you can:

In addition to the base PgBouncer configuration options, set the following values in your pgbouncer.ini file:

The praefect user and its password should be included in the file (default is userlist.txt) used by PgBouncer if the auth_file configuration option is set.

NOTE: By default PgBouncer uses port 6432 to accept incoming connections. You can change it by setting the listen_port configuration option. We recommend setting it to the default port value (5432) used by PostgreSQL instances. Otherwise you should change the configuration parameter praefect['database_port'] for each Praefect instance to the correct value.

Praefect

Introduced in GitLab 13.4, Praefect nodes can no longer be designated as primary.

NOTE: If there are multiple Praefect nodes, complete these steps for each node.

To complete this section you need a configured PostgreSQL server, including:

  • IP/host address (POSTGRESQL_SERVER_ADDRESS)
  • Password (PRAEFECT_SQL_PASSWORD)

Praefect should be run on a dedicated node. Do not run Praefect on the application server, or a Gitaly node.

  1. SSH into the Praefect node and login as root:

    sudo -i
  2. Disable all other services by editing /etc/gitlab/gitlab.rb:

    # Disable all other services on the Praefect node
    postgresql['enable'] = false
    redis['enable'] = false
    nginx['enable'] = false
    alertmanager['enable'] = false
    prometheus['enable'] = false
    grafana['enable'] = false
    puma['enable'] = false
    sidekiq['enable'] = false
    gitlab_workhorse['enable'] = false
    gitaly['enable'] = false
    
    # Enable only the Praefect service
    praefect['enable'] = true
    
    # Prevent database connections during 'gitlab-ctl reconfigure'
    gitlab_rails['rake_cache_clear'] = false
    gitlab_rails['auto_migrate'] = false
  3. Configure Praefect to listen on network interfaces by editing /etc/gitlab/gitlab.rb:

    praefect['listen_addr'] = '0.0.0.0:2305'
    
    # Enable Prometheus metrics access to Praefect. You must use firewalls
    # to restrict access to this address/port.
    praefect['prometheus_listen_addr'] = '0.0.0.0:9652'
  4. Configure a strong auth_token for Praefect by editing /etc/gitlab/gitlab.rb. This is needed by clients outside the cluster (like GitLab Shell) to communicate with the Praefect cluster:

    praefect['auth_token'] = 'PRAEFECT_EXTERNAL_TOKEN'
  5. Configure Praefect to connect to the PostgreSQL database by editing /etc/gitlab/gitlab.rb.

    You need to replace POSTGRESQL_SERVER_ADDRESS with the IP/host address of the database, and PRAEFECT_SQL_PASSWORD with the strong password set above.

    praefect['database_host'] = 'POSTGRESQL_SERVER_ADDRESS'
    praefect['database_port'] = 5432
    praefect['database_user'] = 'praefect'
    praefect['database_password'] = 'PRAEFECT_SQL_PASSWORD'
    praefect['database_dbname'] = 'praefect_production'
    praefect['database_host_no_proxy'] = 'POSTGRESQL_SERVER_ADDRESS'
    praefect['database_port_no_proxy'] = 5432

    If you want to use a TLS client certificate, the options below can be used:

    # Connect to PostgreSQL using a TLS client certificate
    # praefect['database_sslcert'] = '/path/to/client-cert'
    # praefect['database_sslkey'] = '/path/to/client-key'
    
    # Trust a custom certificate authority
    # praefect['database_sslrootcert'] = '/path/to/rootcert'

    By default, Praefect refuses to make an unencrypted connection to PostgreSQL. You can override this by uncommenting the following line:

    # praefect['database_sslmode'] = 'disable'
  6. Configure the Praefect cluster to connect to each Gitaly node in the cluster by editing /etc/gitlab/gitlab.rb.

    The virtual storage's name must match the configured storage name in GitLab configuration. In a later step, we configure the storage name as default so we use default here as well. This cluster has three Gitaly nodes gitaly-1, gitaly-2, and gitaly-3, which are intended to be replicas of each other.

    WARNING: If you have data on an already existing storage called default, you should configure the virtual storage with another name and migrate the data to the Gitaly Cluster storage afterwards.

    Replace PRAEFECT_INTERNAL_TOKEN with a strong secret, which is used by Praefect when communicating with Gitaly nodes in the cluster. This token is distinct from the PRAEFECT_EXTERNAL_TOKEN.

    Replace GITALY_HOST_* with the IP or host address of the each Gitaly node.

    More Gitaly nodes can be added to the cluster to increase the number of replicas. More clusters can also be added for very large GitLab instances.

    NOTE: When adding additional Gitaly nodes to a virtual storage, all storage names within that virtual storage must be unique. Additionally, all Gitaly node addresses referenced in the Praefect configuration must be unique.

    # Name of storage hash must match storage name in git_data_dirs on GitLab
    # server ('default') and in git_data_dirs on Gitaly nodes ('gitaly-1')
    praefect['virtual_storages'] = {
      'default' => {
        'nodes' => {
          'gitaly-1' => {
            'address' => 'tcp://GITALY_HOST_1:8075',
            'token'   => 'PRAEFECT_INTERNAL_TOKEN',
          },
          'gitaly-2' => {
            'address' => 'tcp://GITALY_HOST_2:8075',
            'token'   => 'PRAEFECT_INTERNAL_TOKEN'
          },
          'gitaly-3' => {
            'address' => 'tcp://GITALY_HOST_3:8075',
            'token'   => 'PRAEFECT_INTERNAL_TOKEN'
          }
        }
      }
    }

    NOTE: In GitLab 13.8 and earlier, Gitaly nodes were configured directly under the virtual storage, and not under the nodes key.

  7. Introduced in GitLab 13.1 and later, enable distribution of reads.

  8. Save the changes to /etc/gitlab/gitlab.rb and reconfigure Praefect:

    gitlab-ctl reconfigure
  9. To ensure that Praefect has updated its Prometheus listen address, restart Praefect:

    gitlab-ctl restart praefect
  10. Verify that Praefect can reach PostgreSQL:

    sudo -u git /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml sql-ping

    If the check fails, make sure you have followed the steps correctly. If you edit /etc/gitlab/gitlab.rb, remember to run sudo gitlab-ctl reconfigure again before trying the sql-ping command.

The steps above must be completed for each Praefect node!

Enabling TLS support

Introduced in GitLab 13.2.

Praefect supports TLS encryption. To communicate with a Praefect instance that listens for secure connections, you must:

  • Use a tls:// URL scheme in the gitaly_address of the corresponding storage entry in the GitLab configuration.
  • Bring your own certificates because this isn't provided automatically. The certificate corresponding to each Praefect server must be installed on that Praefect server.

Additionally the certificate, or its certificate authority, must be installed on all Gitaly servers and on all Praefect clients that communicate with it following the procedure described in GitLab custom certificate configuration (and repeated below).

Note the following:

  • The certificate must specify the address you use to access the Praefect server. If addressing the Praefect server by:

    • Hostname, you can either use the Common Name field for this, or add it as a Subject Alternative Name.
    • IP address, you must add it as a Subject Alternative Name to the certificate.
  • You can configure Praefect servers with both an unencrypted listening address listen_addr and an encrypted listening address tls_listen_addr at the same time. This allows you to do a gradual transition from unencrypted to encrypted traffic, if necessary.

To configure Praefect with TLS:

For Omnibus GitLab

  1. Create certificates for Praefect servers.

  2. On the Praefect servers, create the /etc/gitlab/ssl directory and copy your key and certificate there:

    sudo mkdir -p /etc/gitlab/ssl
    sudo chmod 755 /etc/gitlab/ssl
    sudo cp key.pem cert.pem /etc/gitlab/ssl/
    sudo chmod 644 key.pem cert.pem
  3. Edit /etc/gitlab/gitlab.rb and add:

    praefect['tls_listen_addr'] = "0.0.0.0:3305"
    praefect['certificate_path'] = "/etc/gitlab/ssl/cert.pem"
    praefect['key_path'] = "/etc/gitlab/ssl/key.pem"
  4. Save the file and reconfigure.

  5. On the Praefect clients (including each Gitaly server), copy the certificates, or their certificate authority, into /etc/gitlab/trusted-certs:

    sudo cp cert.pem /etc/gitlab/trusted-certs/
  6. On the Praefect clients (except Gitaly servers), edit git_data_dirs in /etc/gitlab/gitlab.rb as follows:

    git_data_dirs({
      "default" => {
        "gitaly_address" => 'tls://LOAD_BALANCER_SERVER_ADDRESS:2305',
        "gitaly_token" => 'PRAEFECT_EXTERNAL_TOKEN'
      }
    })
  7. Save the file and reconfigure GitLab.

For installations from source

  1. Create certificates for Praefect servers.

  2. On the Praefect servers, create the /etc/gitlab/ssl directory and copy your key and certificate there:

    sudo mkdir -p /etc/gitlab/ssl
    sudo chmod 755 /etc/gitlab/ssl
    sudo cp key.pem cert.pem /etc/gitlab/ssl/
    sudo chmod 644 key.pem cert.pem
  3. On the Praefect clients (including each Gitaly server), copy the certificates, or their certificate authority, into the system trusted certificates:

    sudo cp cert.pem /usr/local/share/ca-certificates/praefect.crt
    sudo update-ca-certificates
  4. On the Praefect clients (except Gitaly servers), edit storages in /home/git/gitlab/config/gitlab.yml as follows:

    gitlab:
      repositories:
        storages:
          default:
            gitaly_address: tls://LOAD_BALANCER_SERVER_ADDRESS:3305
            path: /some/local/path

    NOTE: /some/local/path should be set to a local folder that exists, however no data is stored in this folder. This requirement is scheduled to be removed when this issue is resolved.

  5. Save the file and restart GitLab.

  6. Copy all Praefect server certificates, or their certificate authority, to the system trusted certificates on each Gitaly server so the Praefect server trusts the certificate when called by Gitaly servers:

    sudo cp cert.pem /usr/local/share/ca-certificates/praefect.crt
    sudo update-ca-certificates
  7. Edit /home/git/praefect/config.toml and add:

    tls_listen_addr = '0.0.0.0:3305'
    
    [tls]
    certificate_path = '/etc/gitlab/ssl/cert.pem'
    key_path = '/etc/gitlab/ssl/key.pem'
  8. Save the file and restart GitLab.

Gitaly

NOTE: Complete these steps for each Gitaly node.

To complete this section you need:

  • Configured Praefect node
  • 3 (or more) servers, with GitLab installed, to be configured as Gitaly nodes. These should be dedicated nodes, do not run other services on these nodes.

Every Gitaly server assigned to the Praefect cluster needs to be configured. The configuration is the same as a normal standalone Gitaly server, except:

  • The storage names are exposed to Praefect, not GitLab
  • The secret token is shared with Praefect, not GitLab

The configuration of all Gitaly nodes in the Praefect cluster can be identical, because we rely on Praefect to route operations correctly.

Particular attention should be shown to:

  • The gitaly['auth_token'] configured in this section must match the token value under praefect['virtual_storages']['nodes'] on the Praefect node. This was set in the previous section. This document uses the placeholder PRAEFECT_INTERNAL_TOKEN throughout.
  • The storage names in git_data_dirs configured in this section must match the storage names under praefect['virtual_storages'] on the Praefect node. This was set in the previous section. This document uses gitaly-1, gitaly-2, and gitaly-3 as Gitaly storage names.

For more information on Gitaly server configuration, see our Gitaly documentation.

  1. SSH into the Gitaly node and login as root:

    sudo -i
  2. Disable all other services by editing /etc/gitlab/gitlab.rb:

    # Disable all other services on the Praefect node
    postgresql['enable'] = false
    redis['enable'] = false
    nginx['enable'] = false
    grafana['enable'] = false
    puma['enable'] = false
    sidekiq['enable'] = false
    gitlab_workhorse['enable'] = false
    prometheus_monitoring['enable'] = false
    
    # Enable only the Gitaly service
    gitaly['enable'] = true
    
    # Enable Prometheus if needed
    prometheus['enable'] = true
    
    # Prevent database connections during 'gitlab-ctl reconfigure'
    gitlab_rails['rake_cache_clear'] = false
    gitlab_rails['auto_migrate'] = false
  3. Configure Gitaly to listen on network interfaces by editing /etc/gitlab/gitlab.rb:

    # Make Gitaly accept connections on all network interfaces.
    # Use firewalls to restrict access to this address/port.
    gitaly['listen_addr'] = '0.0.0.0:8075'
    
    # Enable Prometheus metrics access to Gitaly. You must use firewalls
    # to restrict access to this address/port.
    gitaly['prometheus_listen_addr'] = '0.0.0.0:9236'
  4. Configure a strong auth_token for Gitaly by editing /etc/gitlab/gitlab.rb. This is needed by clients to communicate with this Gitaly nodes. Typically, this token is the same for all Gitaly nodes.

    gitaly['auth_token'] = 'PRAEFECT_INTERNAL_TOKEN'
  5. Configure the GitLab Shell secret token, which is needed for git push operations. Either:

    • Method 1:

      1. Copy /etc/gitlab/gitlab-secrets.json from the Gitaly client to same path on the Gitaly servers and any other Gitaly clients.
      2. Reconfigure GitLab on Gitaly servers.
    • Method 2:

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

      2. Replace GITLAB_SHELL_SECRET_TOKEN with the real secret.

        gitlab_shell['secret_token'] = 'GITLAB_SHELL_SECRET_TOKEN'
  6. Configure and internal_api_url, which is also needed for git push operations:

    # Configure the gitlab-shell API callback URL. Without this, `git push` will
    # fail. This can be your front door GitLab URL or an internal load balancer.
    # Examples: 'https://gitlab.example.com', 'http://1.2.3.4'
    gitlab_rails['internal_api_url'] = 'http://GITLAB_HOST'
  7. Configure the storage location for Git data by setting git_data_dirs in /etc/gitlab/gitlab.rb. Each Gitaly node should have a unique storage name (such as gitaly-1).

    Instead of configuring git_data_dirs uniquely for each Gitaly node, it is often easier to have include the configuration for all Gitaly nodes on every Gitaly node. This is supported because the Praefect virtual_storages configuration maps each storage name (such as gitaly-1) to a specific node, and requests are routed accordingly. This means every Gitaly node in your fleet can share the same configuration.

    # You can include the data dirs for all nodes in the same config, because
    # Praefect will only route requests according to the addresses provided in the
    # prior step.
    git_data_dirs({
      "gitaly-1" => {
        "path" => "/var/opt/gitlab/git-data"
      },
      "gitaly-2" => {
        "path" => "/var/opt/gitlab/git-data"
      },
      "gitaly-3" => {
        "path" => "/var/opt/gitlab/git-data"
      }
    })
  8. Save the changes to /etc/gitlab/gitlab.rb and reconfigure Gitaly:

    gitlab-ctl reconfigure
  9. To ensure that Gitaly has updated its Prometheus listen address, restart Gitaly:

    gitlab-ctl restart gitaly

The steps above must be completed for each Gitaly node!

After all Gitaly nodes are configured, you can run the Praefect connection checker to verify Praefect can connect to all Gitaly servers in the Praefect configuration.

  1. SSH into each Praefect node and run the Praefect connection checker:

    sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml dial-nodes

Load Balancer

In a fault-tolerant Gitaly configuration, a load balancer is needed to route internal traffic from the GitLab application to the Praefect nodes. The specifics on which load balancer to use or the exact configuration is beyond the scope of the GitLab documentation.

NOTE: The load balancer must be configured to accept traffic from the Gitaly nodes in addition to the GitLab nodes. Some requests handled by gitaly-ruby sidecar processes call into the main Gitaly process. gitaly-ruby uses the Gitaly address set in the GitLab server's git_data_dirs setting to make this connection.

We hope that if you're managing fault-tolerant systems like GitLab, you have a load balancer of choice already. Some examples include HAProxy (open-source), Google Internal Load Balancer, AWS Elastic Load Balancer, F5 Big-IP LTM, and Citrix Net Scaler. This documentation outlines what ports and protocols you need configure.

LB Port Backend Port Protocol
2305 2305 TCP

GitLab

To complete this section you need:

The Praefect cluster needs to be exposed as a storage location to the GitLab application. This is done by updating the git_data_dirs.

Particular attention should be shown to:

  • the storage name added to git_data_dirs in this section must match the storage name under praefect['virtual_storages'] on the Praefect node(s). This was set in the Praefect section of this guide. This document uses default as the Praefect storage name.
  1. SSH into the GitLab node and login as root:

    sudo -i
  2. Configure the external_url so that files could be served by GitLab by proper endpoint access by editing /etc/gitlab/gitlab.rb:

    You need to replace GITLAB_SERVER_URL with the real external facing URL on which current GitLab instance is serving:

    external_url 'GITLAB_SERVER_URL'
  3. Disable the default Gitaly service running on the GitLab host. It isn't needed because GitLab connects to the configured cluster.

    WARNING: If you have existing data stored on the default Gitaly storage, you should migrate the data your Gitaly Cluster storage first.

    gitaly['enable'] = false
  4. Add the Praefect cluster as a storage location by editing /etc/gitlab/gitlab.rb.

    You need to replace:

    • LOAD_BALANCER_SERVER_ADDRESS with the IP address or hostname of the load balancer.
    • PRAEFECT_EXTERNAL_TOKEN with the real secret

    If you are using TLS, the gitaly_address should begin with tls://.

    git_data_dirs({
      "default" => {
        "gitaly_address" => "tcp://LOAD_BALANCER_SERVER_ADDRESS:2305",
        "gitaly_token" => 'PRAEFECT_EXTERNAL_TOKEN'
      }
    })
  5. Configure the GitLab Shell secret token so that callbacks from Gitaly nodes during a git push are properly authenticated. Either:

    • Method 1:

      1. Copy /etc/gitlab/gitlab-secrets.json from the Gitaly client to same path on the Gitaly servers and any other Gitaly clients.
      2. Reconfigure GitLab on Gitaly servers.
    • Method 2:

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

      2. Replace GITLAB_SHELL_SECRET_TOKEN with the real secret.

        gitlab_shell['secret_token'] = 'GITLAB_SHELL_SECRET_TOKEN'
  6. Add Prometheus monitoring settings by editing /etc/gitlab/gitlab.rb. If Prometheus is enabled on a different node, make edits on that node instead.

    You need to replace:

    • PRAEFECT_HOST with the IP address or hostname of the Praefect node
    • GITALY_HOST_* with the IP address or hostname of each Gitaly node
    prometheus['scrape_configs'] = [
      {
        'job_name' => 'praefect',
        'static_configs' => [
          'targets' => [
            'PRAEFECT_HOST:9652', # praefect-1
            'PRAEFECT_HOST:9652', # praefect-2
            'PRAEFECT_HOST:9652', # praefect-3
          ]
        ]
      },
      {
        'job_name' => 'praefect-gitaly',
        'static_configs' => [
          'targets' => [
            'GITALY_HOST_1:9236', # gitaly-1
            'GITALY_HOST_2:9236', # gitaly-2
            'GITALY_HOST_3:9236', # gitaly-3
          ]
        ]
      }
    ]
  7. Save the changes to /etc/gitlab/gitlab.rb and reconfigure GitLab:

    gitlab-ctl reconfigure
  8. Verify on each Gitaly node the Git Hooks can reach GitLab. On each Gitaly node run:

    /opt/gitlab/embedded/bin/gitaly-hooks check /var/opt/gitlab/gitaly/config.toml
  9. Verify that GitLab can reach Praefect:

    gitlab-rake gitlab:gitaly:check
  10. Check in Admin Area > Settings > Repository > Repository storage that the Praefect storage is configured to store new repositories. Following this guide, the default storage should have weight 100 to store all new repositories.

  11. Verify everything is working by creating a new project. Check the "Initialize repository with a README" box so that there is content in the repository that viewed. If the project is created, and you can see the README file, it works!

Use TCP for existing GitLab instances

When adding Gitaly Cluster to an existing Gitaly instance, the existing Gitaly storage must use a TCP address. If gitaly_address is not specified, then a Unix socket is used, which will prevent the communication with the cluster.

For example:

git_data_dirs({
  'default' => { 'gitaly_address' => 'tcp://old-gitaly.internal:8075' },
  'cluster' => {
    'gitaly_address' => 'tcp://<load_balancer_server_address>:2305',
    'gitaly_token' => '<praefect_external_token>'
  }
})

See Mixed Configuration for further information on running multiple Gitaly storages.

Grafana

Grafana is included with GitLab, and can be used to monitor your Praefect cluster. See Grafana Dashboard Service for detailed documentation.

To get started quickly:

  1. SSH into the GitLab node (or whichever node has Grafana enabled) and login as root:

    sudo -i
  2. Enable the Grafana login form by editing /etc/gitlab/gitlab.rb.

    grafana['disable_login_form'] = false
  3. Save the changes to /etc/gitlab/gitlab.rb and reconfigure GitLab:

    gitlab-ctl reconfigure
  4. Set the Grafana administrator password. This command prompts you to enter a new password:

    gitlab-ctl set-grafana-password
  5. In your web browser, open /-/grafana (such as https://gitlab.example.com/-/grafana) on your GitLab server.

    Login using the password you set, and the username admin.

  6. Go to Explore and query gitlab_build_info to verify that you are getting metrics from all your machines.

Congratulations! You've configured an observable fault-tolerant Praefect cluster.

Distributed reads

Praefect supports distribution of read operations across Gitaly nodes that are configured for the virtual node.

All RPCs marked with ACCESSOR option like GetBlob are redirected to an up to date and healthy Gitaly node.

Up to date in this context means that:

  • There is no replication operations scheduled for this node.
  • The last replication operation is in completed state.

If there is no such nodes, or any other error occurs during node selection, the primary node is chosen to serve the request.

To track distribution of read operations, you can use the gitaly_praefect_read_distribution Prometheus counter metric. It has two labels:

  • virtual_storage.
  • storage.

They reflect configuration defined for this instance of Praefect.

Strong consistency

  • Introduced in GitLab 13.1 in alpha, disabled by default.
  • Entered beta in GitLab 13.2, disabled by default.
  • In GitLab 13.3, disabled unless primary-wins voting strategy is disabled.
  • From GitLab 13.4, enabled by default.
  • From GitLab 13.5, you must use Git v2.28.0 or higher on Gitaly nodes to enable strong consistency.
  • From GitLab 13.6, primary-wins voting strategy and gitaly_reference_transactions_primary_wins feature flag were removed from the source code.

Praefect guarantees eventual consistency by replicating all writes to secondary nodes after the write to the primary Gitaly node has happened.

Praefect can instead provide strong consistency by creating a transaction and writing changes to all Gitaly nodes at once. If enabled, transactions are only available for a subset of RPCs. For more information, see the strong consistency epic.

To enable strong consistency:

  • In GitLab 13.5, you must use Git v2.28.0 or higher on Gitaly nodes to enable strong consistency.
  • In GitLab 13.4 and later, the strong consistency voting strategy has been improved and enabled by default. Instead of requiring all nodes to agree, only the primary and half of the secondaries need to agree.
  • In GitLab 13.3, reference transactions are enabled by default with a primary-wins strategy. This strategy causes all transactions to succeed for the primary and thus does not ensure strong consistency. To enable strong consistency, disable the :gitaly_reference_transactions_primary_wins feature flag.
  • In GitLab 13.2, enable the :gitaly_reference_transactions feature flag.
  • In GitLab 13.1, enable the :gitaly_reference_transactions and :gitaly_hooks_rpc feature flags.

Changing feature flags requires access to the Rails console. In the Rails console, enable or disable the flags as required. For example:

Feature.enable(:gitaly_reference_transactions)
Feature.disable(:gitaly_reference_transactions_primary_wins)

To monitor strong consistency, you can use the following Prometheus metrics:

  • gitaly_praefect_transactions_total: Number of transactions created and voted on.
  • gitaly_praefect_subtransactions_per_transaction_total: Number of times nodes cast a vote for a single transaction. This can happen multiple times if multiple references are getting updated in a single transaction.
  • gitaly_praefect_voters_per_transaction_total: Number of Gitaly nodes taking part in a transaction.
  • gitaly_praefect_transactions_delay_seconds: Server-side delay introduced by waiting for the transaction to be committed.
  • gitaly_hook_transaction_voting_delay_seconds: Client-side delay introduced by waiting for the transaction to be committed.

Replication factor

Replication factor is the number of copies Praefect maintains of a given repository. A higher replication factor offers better redundancy and distribution of read workload, but also results in a higher storage cost. By default, Praefect replicates repositories to every storage in a virtual storage.

Configure replication factor

WARNING: Configurable replication factors require repository-specific primary nodes to be used.

Praefect supports configuring a replication factor on a per-repository basis, by assigning specific storage nodes to host a repository.

Praefect does not store the actual replication factor, but assigns enough storages to host the repository so the desired replication factor is met. If a storage node is later removed from the virtual storage, the replication factor of repositories assigned to the storage is decreased accordingly.

You can configure:

  • A default replication factor for each virtual storage that is applied to newly-created repositories. The configuration is added to the /etc/gitlab/gitlab.rb file:

    praefect['virtual_storages'] = {
      'default' => {
        'default_replication_factor' => 1,
        # ...
      }
    }
  • A replication factor for an existing repository using the set-replication-factor sub-command. set-replication-factor automatically assigns or unassigns random storage nodes as necessary to reach the desired replication factor. The repository's primary node is always assigned first and is never unassigned.

    sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml set-replication-factor -virtual-storage <virtual-storage> -repository <relative-path> -replication-factor <replication-factor>
    • -virtual-storage is the virtual storage the repository is located in.
    • -repository is the repository's relative path in the storage.
    • -replication-factor is the desired replication factor of the repository. The minimum value is 1, as the primary needs a copy of the repository. The maximum replication factor is the number of storages in the virtual storage.

    On success, the assigned host storages are printed. For example:

    $ sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml set-replication-factor -virtual-storage default -repository @hashed/3f/db/3fdba35f04dc8c462986c992bcf875546257113072a909c162f7e470e581e278.git -replication-factor 2
    
    current assignments: gitaly-1, gitaly-2

Automatic failover and primary election strategies

Praefect regularly checks the health of each Gitaly node. This is used to automatically fail over to a newly-elected primary Gitaly node if the current primary node is found to be unhealthy.

We recommend using repository-specific primary nodes. This is planned to be the only available election strategy from GitLab 14.0.

Repository-specific primary nodes

Introduced in GitLab 13.12.

Gitaly Cluster supports electing repository-specific primary Gitaly nodes. Repository-specific Gitaly primary nodes are enabled in /etc/gitlab/gitlab.rb by setting praefect['failover_election_strategy'] = 'per_repository'.

Praefect's deprecated election strategies:

  • Elected a primary Gitaly node for each virtual storage, which was used as the primary node for each repository in the virtual storage.
  • Prevented horizontal scaling of a virtual storage. The primary Gitaly node needed a replica of each repository and thus became the bottleneck.

The per_repository election strategy solves this problem by electing a primary Gitaly node separately for each repository. Combined with configurable replication factors, you can horizontally scale storage capacity and distribute write load across Gitaly nodes.

Primary elections are run when:

  • Praefect starts up.
  • The cluster's consensus of a Gitaly node's health changes.

A Gitaly node is considered:

  • Healthy if >=50% Praefect nodes have successfully health checked the Gitaly node in the previous ten seconds.
  • Unhealthy otherwise.

During an election run, Praefect elects a new primary Gitaly node for each repository that has an unhealthy primary Gitaly node. The election is made:

  • Randomly from healthy secondary Gitaly nodes that are the most up to date.
  • Only from Gitaly nodes assigned to the host repository.

If there are no healthy secondary nodes for a repository:

  • The unhealthy primary node is demoted and the repository is left without a primary node.
  • Operations that require a primary node fail until a primary is successfully elected.

Migrate to repository-specific primary Gitaly nodes

New Gitaly Clusters can start using the per_repository election strategy immediately.

To migrate existing clusters:

  1. Praefect nodes didn't historically keep database records of every repository stored on the cluster. When the per_repository election strategy is configured, Praefect expects to have database records of each repository. A background migration is included in GitLab 13.6 and later to create any missing database records for repositories. Before migrating you should verify the migration has run by checking Praefect's logs:

    Check Praefect's logs for repository importer finished message. The virtual_storages field contains the names of virtual storages and whether they've had any missing database records created.

    For example, the default virtual storage has been successfully migrated:

    {"level":"info","msg":"repository importer finished","pid":19752,"time":"2021-04-28T11:41:36.743Z","virtual_storages":{"default":true}}

    If a virtual storage has not been successfully migrated, it would have false next to it:

    {"level":"info","msg":"repository importer finished","pid":19752,"time":"2021-04-28T11:41:36.743Z","virtual_storages":{"default":false}}

    The migration is ran when Praefect starts up. If the migration is unsuccessful, you can restart a Praefect node to reattempt it. The migration only runs with sql election strategy configured.

  2. Running two different election strategies side by side can cause a split brain, where different Praefect nodes consider repositories to have different primaries. To avoid this, shut down all Praefect nodes before changing the election strategy.

    Do this by running gitlab-ctl stop praefect on the Praefect nodes.

  3. On the Praefect nodes, configure the election strategy in /etc/gitlab/gitlab.rb with praefect['failover_election_strategy'] = 'per_repository'.

  4. Finally, run gitlab-ctl reconfigure to reconfigure and restart the Praefect nodes.

Deprecated election strategies

WARNING: The below election strategies are deprecated and are scheduled for removal in GitLab 14.0. Migrate to repository-specific primary nodes.

  • PostgreSQL: Enabled by default until GitLab 14.0, and equivalent to: praefect['failover_election_strategy'] = 'sql'.

    This configuration option:

    • Allows multiple Praefect nodes to coordinate via the PostgreSQL database to elect a primary Gitaly node.
    • Causes Praefect nodes to elect a new primary Gitaly node, monitor its health, and elect a new primary Gitaly node if the current one is not reached within 10 seconds by a majority of the Praefect nodes.
  • Memory: Enabled by setting praefect['failover_election_strategy'] = 'local' in /etc/gitlab/gitlab.rb on the Praefect node.

    If a sufficient number of health checks fail for the current primary Gitaly node, a new primary is elected. Do not use with multiple Praefect nodes! Using with multiple Praefect nodes is likely to result in a split brain.

Primary Node Failure

Gitaly Cluster recovers from a failing primary Gitaly node by promoting a healthy secondary as the new primary.

To minimize data loss, Gitaly Cluster:

  • Switches repositories that are outdated on the new primary to read-only mode.
  • Elects the secondary with the least unreplicated writes from the primary to be the new primary. Because there can still be some unreplicated writes, data loss can occur.

Read-only mode

  • Introduced in GitLab 13.0 as generally available.
  • Between GitLab 13.0 and GitLab 13.2, read-only mode applied to the whole virtual storage and occurred whenever failover occurred.
  • In GitLab 13.3 and later, read-only mode applies on a per-repository basis and only occurs if a new primary is out of date.

When Gitaly Cluster switches to a new primary, repositories enter read-only mode if they are out of date. This can happen after failing over to an outdated secondary. Read-only mode eases data recovery efforts by preventing writes that may conflict with the unreplicated writes on other nodes.

To enable writes again, an administrator can:

  1. Check for data loss.
  2. Attempt to recover missing data.
  3. Either enable writes in the virtual storage or accept data loss if necessary, depending on the version of GitLab.

Check for data loss

The Praefect dataloss sub-command identifies replicas that are likely to be outdated. This can help identify potential data loss after a failover. The following parameters are available:

  • -virtual-storage that specifies which virtual storage to check. The default behavior is to display outdated replicas of read-only repositories as they might require administrator action.
  • In GitLab 13.3 and later, -partially-replicated that specifies whether to display a list of outdated replicas of writable repositories.

NOTE: dataloss is still in beta and the output format is subject to change.

To check for repositories with outdated primaries, run:

sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml dataloss [-virtual-storage <virtual-storage>]

Every configured virtual storage is checked if none is specified:

sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml dataloss

Repositories which have assigned storage nodes that contain an outdated copy of the repository are listed in the output. This information is printed for each repository:

  • A repository's relative path to the storage directory identifies each repository and groups the related information.
  • The repository's current status is printed in parentheses next to the disk path. If the repository's primary is outdated, the repository is in read-only mode and can't accept writes. Otherwise, the mode is writable.
  • The primary field lists the repository's current primary. If the repository has no primary, the field shows No Primary.
  • The In-Sync Storages lists replicas which have replicated the latest successful write and all writes preceding it.
  • The Outdated Storages lists replicas which contain an outdated copy of the repository. Replicas which have no copy of the repository but should contain it are also listed here. The maximum number of changes the replica is missing is listed next to replica. It's important to notice that the outdated replicas may be fully up to date or contain later changes but Praefect can't guarantee it.

Whether a replica is assigned to host the repository is listed with each replica's status. assigned host is printed next to replicas which are assigned to store the repository. The text is omitted if the replica contains a copy of the repository but is not assigned to store the repository. Such replicas aren't kept in-sync by Praefect, but may act as replication sources to bring assigned replicas up to date.

Example output:

Virtual storage: default
  Outdated repositories:
    @hashed/3f/db/3fdba35f04dc8c462986c992bcf875546257113072a909c162f7e470e581e278.git (read-only):
      Primary: gitaly-1
      In-Sync Storages:
        gitaly-2, assigned host
      Outdated Storages:
        gitaly-1 is behind by 3 changes or less, assigned host
        gitaly-3 is behind by 3 changes or less

A confirmation is printed out when every repository is writable. For example:

Virtual storage: default
  All repositories are writable!

Outdated replicas of writable repositories

Introduced in GitLab 13.3.

To also list information of repositories whose primary is up to date but one or more assigned replicas are outdated, use the -partially-replicated flag.

A repository is writable if the primary has the latest changes. Secondaries might be temporarily outdated while they are waiting to replicate the latest changes.

sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml dataloss [-virtual-storage <virtual-storage>] [-partially-replicated]

Example output:

Virtual storage: default
  Outdated repositories:
    @hashed/3f/db/3fdba35f04dc8c462986c992bcf875546257113072a909c162f7e470e581e278.git (writable):
      Primary: gitaly-1
      In-Sync Storages:
        gitaly-1, assigned host
      Outdated Storages:
        gitaly-2 is behind by 3 changes or less, assigned host
        gitaly-3 is behind by 3 changes or less

With the -partially-replicated flag set, a confirmation is printed out if every assigned replica is fully up to date.

For example:

Virtual storage: default
  All repositories are up to date!

Check repository checksums

To check a project's repository checksums across on all Gitaly nodes, run the replicas Rake task on the main GitLab node.

Enable writes or accept data loss

Praefect provides the following sub-commands to re-enable writes:

  • In GitLab 13.2 and earlier, enable-writes to re-enable virtual storage for writes after data recovery attempts.

    sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml enable-writes -virtual-storage <virtual-storage>
  • In GitLab 13.3 and later, accept-dataloss to accept data loss and re-enable writes for repositories after data recovery attempts have failed. Accepting data loss causes current version of the repository on the authoritative storage to be considered latest. Other storages are brought up to date with the authoritative storage by scheduling replication jobs.

    sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml accept-dataloss -virtual-storage <virtual-storage> -repository <relative-path> -authoritative-storage <storage-name>

WARNING: accept-dataloss causes permanent data loss by overwriting other versions of the repository. Data recovery efforts must be performed before using it.

Data recovery

If a Gitaly node fails replication jobs for any reason, it ends up hosting outdated versions of the affected repositories. Praefect provides tools for:

  • Automatic reconciliation, for GitLab 13.4 and later.
  • Manual reconciliation, for:
    • GitLab 13.3 and earlier.
    • Repositories upgraded to GitLab 13.4 and later without entries in the repositories table. In GitLab 13.6 and later, a migration is run when Praefect starts for these repositories.

These tools reconcile the outdated repositories to bring them fully up to date again.

Automatic reconciliation

Introduced in GitLab 13.4.

Praefect automatically reconciles repositories that are not up to date. By default, this is done every five minutes. For each outdated repository on a healthy Gitaly node, the Praefect picks a random, fully up-to-date replica of the repository on another healthy Gitaly node to replicate from. A replication job is scheduled only if there are no other replication jobs pending for the target repository.

The reconciliation frequency can be changed via the configuration. The value can be any valid Go duration value. Values below 0 disable the feature.

Examples:

praefect['reconciliation_scheduling_interval'] = '5m' # the default value
praefect['reconciliation_scheduling_interval'] = '30s' # reconcile every 30 seconds
praefect['reconciliation_scheduling_interval'] = '0' # disable the feature

Manual reconciliation

WARNING: The reconcile sub-command is deprecated and scheduled for removal in GitLab 14.0. Use automatic reconciliation instead. Manual reconciliation may produce excess replication jobs and is limited in functionality. Manual reconciliation does not work when repository-specific primary nodes are enabled.

The Praefect reconcile sub-command allows for the manual reconciliation between two Gitaly nodes. The command replicates every repository on a later version on the reference storage to the target storage.

sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml reconcile -virtual <virtual-storage> -reference <up-to-date-storage> -target <outdated-storage> -f
  • Replace the placeholder <virtual-storage> with the virtual storage containing the Gitaly node storage to be checked.
  • Replace the placeholder <up-to-date-storage> with the Gitaly storage name containing up to date repositories.
  • Replace the placeholder <outdated-storage> with the Gitaly storage name containing outdated repositories.

Migrate to Gitaly Cluster

Whether migrating to Gitaly Cluster because of NFS support deprecation or to move from single Gitaly nodes, the basic process involves:

  1. Create the required storage.
  2. Create and configure Gitaly Cluster.
  3. Move the repositories.

The size of the required storage can vary between instances and depends on the set replication factor. The migration to Gitaly Cluster might include implementing repository storage redundancy.

For a replication factor:

  • Of 1: NFS, Gitaly, and Gitaly Cluster have roughly the same storage requirements.
  • More than 1: The amount of required storage is used space * replication factor. used space should include any planned future growth.

Move Repositories

To migrate to Gitaly Cluster, existing repositories stored outside Gitaly Cluster must be moved. There is no automatic migration but the moves can be scheduled with the GitLab API.

GitLab repositories can be associated with projects, groups, and snippets. Each of these types have a separate API to schedule the respective repositories to move. To move all repositories on a GitLab instance, each of these types must be scheduled to move for each storage.

Each repository is made read-only for the duration of the move. The repository is not writable until the move has completed.

After creating and configuring Gitaly Cluster:

  1. Ensure all storages are accessible to the GitLab instance. In this example, these are <original_storage_name> and <cluster_storage_name>.
  2. Configure repository storage weights so that the Gitaly Cluster receives all new projects. This stops new projects being created on existing Gitaly nodes while the migration is in progress.
  3. Schedule repository moves for:

Bulk schedule project moves

  1. Schedule repository storage moves for all projects on a storage shard using the API. For example:

    curl --request POST --header "Private-Token: <your_access_token>" --header "Content-Type: application/json" \
    --data '{"source_storage_name":"<original_storage_name>","destination_storage_name":"<cluster_storage_name>"}' "https://gitlab.example.com/api/v4/project_repository_storage_moves"
  2. Query the most recent repository moves using the API. The query indicates either:

    • The moves have completed successfully. The state field is finished.
    • The moves are in progress. Re-query the repository move until it completes successfully.
    • The moves have failed. Most failures are temporary and are solved by rescheduling the move.
  3. After the moves are complete, query projects using the API to confirm that all projects have moved. No projects should be returned with repository_storage field set to the old storage.

    curl --header "Private-Token: <your_access_token>" --header "Content-Type: application/json" \
    "https://gitlab.example.com/api/v4/projects?repository_storage=<original_storage_name>"

    Alternatively use the rails console to confirm that all projects have moved. Run the following in the rails console:

    ProjectRepository.for_repository_storage('<original_storage_name>')
  4. Repeat for each storage as required.

Bulk schedule snippet moves

  1. Schedule repository storage moves for all snippets on a storage shard using the API. For example:

    curl --request POST --header "PRIVATE-TOKEN: <your_access_token>" --header "Content-Type: application/json" \
    --data '{"source_storage_name":"<original_storage_name>","destination_storage_name":"<cluster_storage_name>"}' "https://gitlab.example.com/api/v4/snippet_repository_storage_moves"
  2. Query the most recent repository moves using the API. The query indicates either:

    • The moves have completed successfully. The state field is finished.
    • The moves are in progress. Re-query the repository move until it completes successfully.
    • The moves have failed. Most failures are temporary and are solved by rescheduling the move.
  3. After the moves are complete, use the rails console to confirm that all snippets have moved. No snippets should be returned for the original storage. Run the following in the rails console:

    SnippetRepository.for_repository_storage('<original_storage_name>')
  4. Repeat for each storage as required.

Bulk schedule group moves (PREMIUM SELF)

  1. Schedule repository storage moves for all groups on a storage shard using the API.

    curl --request POST --header "PRIVATE-TOKEN: <your_access_token>" --header "Content-Type: application/json" \
    --data '{"source_storage_name":"<original_storage_name>","destination_storage_name":"<cluster_storage_name>"}' "https://gitlab.example.com/api/v4/group_repository_storage_moves"
  2. Query the most recent repository moves using the API. The query indicates either:

    • The moves have completed successfully. The state field is finished.
    • The moves are in progress. Re-query the repository move until it completes successfully.
    • The moves have failed. Most failures are temporary and are solved by rescheduling the move.
  3. After the moves are complete, use the rails console to confirm that all groups have moved. No groups should be returned for the original storage. Run the following in the rails console:

    GroupWikiRepository.for_repository_storage('<original_storage_name>')
  4. Repeat for each storage as required.

Debugging Praefect

If you receive an error, check /var/log/gitlab/gitlab-rails/production.log.

Here are common errors and potential causes:

  • 500 response code
    • ActionView::Template::Error (7:permission denied)
      • praefect['auth_token'] and gitlab_rails['gitaly_token'] do not match on the GitLab server.
    • Unable to save project. Error: 7:permission denied
      • Secret token in praefect['storage_nodes'] on GitLab server does not match the value in gitaly['auth_token'] on one or more Gitaly servers.
  • 503 response code
    • GRPC::Unavailable (14:failed to connect to all addresses)
      • GitLab was unable to reach Praefect.
    • GRPC::Unavailable (14:all SubCons are in TransientFailure...)
      • Praefect cannot reach one or more of its child Gitaly nodes. Try running the Praefect connection checker to diagnose.

Determine primary Gitaly node

To determine the current primary Gitaly node for a specific Praefect node:

  • Use the Shard Primary Election Grafana chart on the Gitlab Omnibus - Praefect dashboard. This is recommended.

  • If you do not have Grafana set up, use the following command on each host of each Praefect node:

    curl localhost:9652/metrics | grep gitaly_praefect_primaries`