• Using external storage
• Using object storage
  • Object Storage Settings
    • S3 compatible connection settings
• Alternative in-database storage
• Correcting incorrectly-migrated diffs
• Switching from external storage to object storage

Merge request diffs storage

Merge request diffs are size-limited copies of diffs associated with merge requests. When viewing a merge request, diffs are sourced from these copies wherever possible as a performance optimization.

By default, merge request diffs are stored in the database, in a table named merge_request_diff_files. Larger installations may find this table grows too large, in which case, switching to external storage is recommended.

Merge request diffs can be stored on disk, or in object storage. In general, it is better to store the diffs in the database than on disk. A compromise is available that only stores outdated diffs outside of database.

Using external storage

In Omnibus installations:

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

   gitlab_rails['external_diffs_enabled'] = true
   

2. The external diffs are stored in /var/opt/gitlab/gitlab-rails/shared/external-diffs. To change the path, for example, to /mnt/storage/external-diffs, edit /etc/gitlab/gitlab.rb and add the following line:

   gitlab_rails['external_diffs_storage_path'] = "/mnt/storage/external-diffs"
   

3. Save the file and reconfigure GitLab for the changes to take effect. GitLab then migrates your existing merge request diffs to external storage.

In installations from source:

1. Edit /home/git/gitlab/config/gitlab.yml and add or amend the following lines:

   external_diffs:
     enabled: true
   

2. The external diffs are stored in /home/git/gitlab/shared/external-diffs. To change the path, for example, to /mnt/storage/external-diffs, edit /home/git/gitlab/config/gitlab.yml and add or amend the following lines:

   external_diffs:
     enabled: true
     storage_path: /mnt/storage/external-diffs
   

3. Save the file and restart GitLab for the changes to take effect. GitLab then migrates your existing merge request diffs to external storage.

Using object storage

Instead of storing the external diffs on disk, we recommended the use of an object store like AWS S3 instead. This configuration relies on valid AWS credentials to be configured already.

In Omnibus installations:

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

   gitlab_rails['external_diffs_enabled'] = true
   

2. Set object storage settings.
3. Save the file and reconfigure GitLab for the changes to take effect. GitLab then migrates your existing merge request diffs to external storage.

In installations from source:

1. Edit /home/git/gitlab/config/gitlab.yml and add or amend the following lines:

   external_diffs:
     enabled: true
   

2. Set object storage settings.
3. Save the file and restart GitLab for the changes to take effect. GitLab then migrates your existing merge request diffs to external storage.

Read more about using object storage with GitLab.

Object Storage Settings

In GitLab 13.2 and later, you should use the consolidated object storage settings. This section describes the earlier configuration format.

For source installations, these settings are nested under external_diffs: and then object_store:. On Omnibus installations, they are prefixed by external_diffs_object_store_.

S3 compatible connection settings

See the available connection settings for different providers.

In Omnibus installations:

1. Edit /etc/gitlab/gitlab.rb and add the following lines by replacing with the values you want:

   gitlab_rails['external_diffs_enabled'] = true
   gitlab_rails['external_diffs_object_store_enabled'] = true
   gitlab_rails['external_diffs_object_store_remote_directory'] = "external-diffs"
   gitlab_rails['external_diffs_object_store_connection'] = {
     'provider' => 'AWS',
     'region' => 'eu-central-1',
     'aws_access_key_id' => 'AWS_ACCESS_KEY_ID',
     'aws_secret_access_key' => 'AWS_SECRET_ACCESS_KEY'
   }
   

   If you are using AWS IAM profiles, omit the AWS access key and secret access key/value pairs. For example:

   gitlab_rails['external_diffs_object_store_connection'] = {
     'provider' => 'AWS',
     'region' => 'eu-central-1',
     'use_iam_profile' => true
   }
   

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

In installations from source:

1. Edit /home/git/gitlab/config/gitlab.yml and add or amend the following lines:

   external_diffs:
     enabled: true
     object_store:
       enabled: true
       remote_directory: "external-diffs" # The bucket name
       connection:
         provider: AWS # Only AWS supported at the moment
         aws_access_key_id: AWS_ACCESS_KEY_ID
         aws_secret_access_key: AWS_SECRET_ACCESS_KEY
         region: eu-central-1
   

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

Alternative in-database storage

Enabling external diffs may reduce the performance of merge requests, as they must be retrieved in a separate operation to other data. A compromise may be reached by only storing outdated diffs externally, while keeping current diffs in the database.

To enable this feature, perform the following steps:

In Omnibus installations:

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

   gitlab_rails['external_diffs_when'] = 'outdated'
   

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

In installations from source:

1. Edit /home/git/gitlab/config/gitlab.yml and add or amend the following lines:

   external_diffs:
     enabled: true
     when: outdated
   

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

With this feature enabled, diffs are initially stored in the database, rather than externally. They are moved to external storage after any of these conditions become true:

• A newer version of the merge request diff exists
• The merge request was merged more than seven days ago
• The merge request was closed more than seven day ago

These rules strike a balance between space and performance by only storing frequently-accessed diffs in the database. Diffs that are less likely to be accessed are moved to external storage instead.

Correcting incorrectly-migrated diffs

Versions of GitLab earlier than v13.0.0 would incorrectly record the location of some merge request diffs when external diffs in object storage were enabled. This mainly affected imported merge requests, and was resolved with this merge request.

If you are using object storage, or have never used on-disk storage for external diffs, the Changes tab for some merge requests fails to load with a 500 error, and the exception for that error is of this form:

Errno::ENOENT (No such file or directory @ rb_sysopen - /var/opt/gitlab/gitlab-rails/shared/external-diffs/merge_request_diffs/mr-6167082/diff-8199789)

Then you are affected by this issue. Because it’s not possible to safely determine all these conditions automatically, we’ve provided a Rake task in GitLab v13.2.0 that you can run manually to correct the data:

In Omnibus installations:

sudo gitlab-rake gitlab:external_diffs:force_object_storage

In installations from source:

sudo -u git -H bundle exec rake gitlab:external_diffs:force_object_storage RAILS_ENV=production

Environment variables can be provided to modify the behavior of the task. The available variables are:

The START_ID and END_ID variables may be used to run the update in parallel, by assigning different processes to different parts of the table. The BATCH and UPDATE_DELAY parameters allow the speed of the migration to be traded off against concurrent access to the table. The ANSI parameter should be set to false if your terminal does not support ANSI escape codes.

By default, sudo does not preserve existing environment variables. You should append them, rather than prefix them.

sudo gitlab-rake gitlab:external_diffs:force_object_storage START_ID=59946109 END_ID=59946109 UPDATE_DELAY=5

Switching from external storage to object storage

Automatic migration moves diffs stored in the database, but it does not move diffs between storage types. To switch from external storage to object storage:

1. Move files stored on local or NFS storage to object storage manually.
2. Run the Rake task in the previous section to change their location in the database.