Recovery options

Gitaly Cluster can recover from certain types of failure.

Primary Node Failure

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

In GitLab 14.1 and later, Gitaly Cluster:

• Elects a healthy secondary with a fully up to date copy of the repository as the new primary.
• Repository becomes unavailable if there are no fully up to date copies of it on healthy secondaries.

To minimize data loss in GitLab 13.0 to 14.0, 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• Switches repositories that are outdated on the new primary to read-only mode.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• In GitLab 13.3 and later, read-only mode applies on a per-repository basis and only occurs if a new primary is out of date. new primary. If the failed primary contained unreplicated writes, data loss can occur.
• Removed in GitLab 14.1. Instead, repositories become unavailable.

When Gitaly Cluster switches to a new primary in GitLab 13.0 to 14.0, repositories enter read-only mode if they are out of date. This can happen after failing over to an outdated secondary.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 in GitLab 13.0 to 14.0, 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.

Unavailable repositories

• From GitLab 13.0 through 14.0, repositories became read-only if they were outdated on the primary but fully up to date on a healthy secondary. dataloss sub-command displays read-only repositories by default through these versions.
• Since GitLab 14.1, Praefect contains more responsive failover logic which immediately fails over to one of the fully up to date secondaries rather than placing the repository in read-only mode. Since GitLab 14.1, the dataloss sub-command displays repositories which are unavailable due to having no fully up to date copies on healthy Gitaly nodes.

A repository is unavailable if all of its up to date replicas are unavailable. Unavailable repositories are not accessible through Praefect to prevent serving stale data that may break automated tooling.

Check for data loss

The Praefect dataloss subcommand identifies:

• Copies of repositories in GitLab 13.0 to GitLab 14.0 that at are likely to be outdated. This can help identify potential data loss after a failover.
• Repositories in GitLab 14.1 and later that are unavailable. This helps identify potential data loss and repositories which are no longer accessible because all of their up-to-date replicas copies are unavailable.

The following parameters are available:

• -virtual-storage that specifies which virtual storage to check. Because they might require an administrator to intervene, the default behavior is to display:
  • In GitLab 13.0 to 14.0, copies of read-only repositories.
  • In GitLab 14.1 and later, unavailable repositories.
• In GitLab 14.1 and later, -partially-unavailable that specifies whether to include in the output repositories that are available but have some assigned copies that are not available.

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

To check for repositories with outdated primaries or for unavailable repositories, 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 are listed in the output that have either:

• An outdated copy of the repository on the primary, in GitLab 13.0 to GitLab 14.0.
• No healthy and fully up-to-date copies available, in GitLab 14.1 and later.

The following 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:
  • In GitLab 13.0 to 14.0, either (read-only) if the repository's primary node is outdated and can't accept writes. Otherwise, (writable).
  • In GitLab 14.1 and later, (unavailable) is printed next to the disk path if the repository is unavailable.
• 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.

Additional information includes:

• Whether a node is assigned to host the repository is listed with each node's status. assigned host is printed next to nodes that are assigned to store the repository. The text is omitted if the node contains a copy of the repository but is not assigned to store the repository. Such copies aren't kept in sync by Praefect, but may act as replication sources to bring assigned copies up to date.
• In GitLab 14.1 and later, unhealthy is printed next to the copies that are located on unhealthy Gitaly nodes.

Example output:

Virtual storage: default
  Outdated repositories:
    @hashed/3f/db/3fdba35f04dc8c462986c992bcf875546257113072a909c162f7e470e581e278.git (unavailable):
      Primary: gitaly-1
      In-Sync Storages:
        gitaly-2, assigned host, unhealthy
      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 available. For example:

Virtual storage: default
  All repositories are available!

Unavailable replicas of available repositories

NOTE: In GitLab 14.0 and earlier, the flag is -partially-replicated and the output shows any repositories with assigned nodes with outdated copies.

To also list information of repositories which are available but are unavailable from some of the assigned nodes, use the -partially-unavailable flag.

A repository is available if there is a healthy, up to date replica available. Some of the assigned secondary replicas may be temporarily unavailable for access 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-unavailable]

Example output:

Virtual storage: default
  Outdated repositories:
    @hashed/3f/db/3fdba35f04dc8c462986c992bcf875546257113072a909c162f7e470e581e278.git:
      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-unavailable flag set, a confirmation is printed out if every assigned replica is fully up to date and healthy.

For example:

Virtual storage: default
  All repositories are fully available on all assigned storages!

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.

Accept data loss

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

If it is not possible to bring one of the up to date replicas back online, you may have to accept data loss. When accepting data loss, Praefect marks the chosen replica of the repository as the latest version and replicates it to the other assigned Gitaly nodes. This process overwrites any other version of the repository so care must be taken.

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>

Enable writes or accept data loss

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

Praefect provides the following subcommands to re-enable writes or accept data loss:

• 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, if it is not possible to bring one of the up to date nodes back online, you may have to accept data loss:

  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>

  When accepting data loss, Praefect:

  1. Marks the chosen copy of the repository as the latest version.
  2. Replicates the copy to the other assigned Gitaly nodes.

  This process overwrites any other copy of the repository so care must be taken.

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 was removed in GitLab 14.1. 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.

Manually remove repositories

• Introduced in GitLab 14.3.
• Introduced in GitLab 14.6, support for dry-run mode.

The remove-repository Praefect sub-command removes a repository from a Gitaly Cluster, and all state associated with a given repository including:

• On-disk repositories on all relevant Gitaly nodes.
• Any database state tracked by Praefect.

In GitLab 14.6 and later, by default, the command operates in dry-run mode. In earlier versions, the command didn't support dry-run mode. For example:

sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml remove-repository -virtual-storage <virtual-storage> -repository <repository>

• Replace <virtual-storage> with the name of the virtual storage containing the repository.

• Replace <repository> with the relative path of the repository to remove.

• In GitLab 14.6 and later, add -apply to run the command outside of dry-run mode and remove the repository. For example:

  sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml remove-repository -virtual-storage <virtual-storage> -repository <repository> -apply

• -virtual-storage is the virtual storage the repository is located in. Virtual storages are configured in /etc/gitlab/gitlab.rb under praefect['virtual_storages] and looks like the following:

  praefect['virtual_storages'] = {
    'default' => {
      ...
    },
    'storage-1' => {
      ...
    }
  }

  In this example, the virtual storage to specify is default or storage-1.

• -repository is the repository's relative path in the storage beginning with @hashed. For example:

  @hashed/f5/ca/f5ca38f748a1d6eaf726b8a42fb575c3c71f1864a8143301782de13da2d9202b.git

Parts of the repository can continue to exist after running remove-repository. This can be because of:

• A deletion error.
• An in-flight RPC call targeting the repository.

If this occurs, run remove-repository again.

Manually list untracked repositories

Introduced in GitLab 14.4.

The list-untracked-repositories Praefect sub-command lists repositories of the Gitaly Cluster that both:

• Exist for at least one Gitaly storage.
• Aren't tracked in the Praefect database.

The command outputs:

• Result to STDOUT and the command's logs.
• Errors to STDERR.

Each entry is a complete JSON string with a newline at the end (configurable using the -delimiter flag). For example:

sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml list-untracked-repositories
{"virtual_storage":"default","storage":"gitaly-1","relative_path":"@hashed/ab/cd/abcd123456789012345678901234567890123456789012345678901234567890.git"}
{"virtual_storage":"default","storage":"gitaly-1","relative_path":"@hashed/ab/cd/abcd123456789012345678901234567890123456789012345678901234567891.git"}

Manually track repositories

• Introduced in GitLab 14.4.
• Introduced in GitLab 14.6, support for immediate replication.

The track-repository Praefect sub-command adds repositories on disk to the Praefect database to be tracked.

sudo /opt/gitlab/embedded/bin/praefect -config /var/opt/gitlab/praefect/config.toml track-repository -virtual-storage <virtual-storage> -authoritative-storage <storage-name> -repository <repository> -replicate-immediately

• -virtual-storage is the virtual storage the repository is located in. Virtual storages are configured in /etc/gitlab/gitlab.rb under praefect['virtual_storages] and looks like the following:

  praefect['virtual_storages'] = {
    'default' => {
      ...
    },
    'storage-1' => {
      ...
    }
  }

  In this example, the virtual storage to specify is default or storage-1.

• -repository is the repository's relative path in the storage beginning with @hashed. For example:

  @hashed/f5/ca/f5ca38f748a1d6eaf726b8a42fb575c3c71f1864a8143301782de13da2d9202b.git

• -authoritative-storage is the storage we want Praefect to treat as the primary. Required if per-repository replication is set as the replication strategy.

• -replicate-immediately, available in GitLab 14.6 and later, causes the command to replicate the repository to its secondaries immediately. Otherwise, replication jobs are scheduled for execution in the database and are picked up by a Praefect background process.

The command outputs:

• Results to STDOUT and the command's logs.
• Errors to STDERR.

This command fails if:

• The repository is already being tracked by the Praefect database.
• The repository does not exist on disk.