Reduce repository size

Git repositories become larger over time. When large files are added to a Git repository:

  • Fetching the repository becomes slower because everyone must download the files.
  • They take up a large amount of storage space on the server.
  • Git repository storage limits can be reached.

Rewriting a repository can remove unwanted history to make the repository smaller. We recommend git filter-repo over git filter-branch and BFG.

DANGER: Warning: Rewriting repository history is a destructive operation. Make sure to back up your repository before you begin. The best way back up a repository is to export the project.

NOTE: Note: Git LFS files can only be removed by an Administrator using a Rake task. Removal of this limitation is planned.

Purge files from repository history

To reduce the size of your repository in GitLab, you must remove references to large files from branches, tags, and other internal references (refs) that are automatically created by GitLab. These refs include:

  • refs/merge-requests/* for merge requests.
  • refs/pipelines/* for pipelines.
  • refs/environments/* for environments.

Git doesn't usually download these refs to make cloning and fetch faster, but we can use the --mirror option to download all the advertised refs.

  1. Install git filter-repo using a supported package manager or from source.

  2. Clone a fresh copy of the repository using --bare and --mirror:

    git clone --bare --mirror https://gitlab.example.com/my/project.git
  3. Using git filter-repo, purge any files from the history of your repository.

    To purge large files, the --strip-blobs-bigger-than option can be used:

    git filter-repo --strip-blobs-bigger-than 10M

    To purge large files stored using Git LFS, the --blob--callback option can be used. The example below, uses the callback to read the file size from the Git LFS pointer, and removes files larger than 10MB.

    git filter-repo --blob-callback '
      if blob.data.startswith(b"version https://git-lfs.github.com/spec/v1"):
        size_in_bytes = int.from_bytes(blob.data[124:], byteorder="big")
        if size_in_bytes > 10*1000:
          blob.skip()
      '

    To purge specific large files by path, the --path and --invert-paths options can be combined:

    git filter-repo --path path/to/big/file.m4v --invert-paths

    See the git filter-repo documentation for more examples and the complete documentation.

  4. Force push your changes to overwrite all branches on GitLab:

    git push origin --force 'refs/heads/*'

    Protected branches will cause this to fail. To proceed, you must remove branch protection, push, and then re-enable protected branches.

  5. To remove large files from tagged releases, force push your changes to all tags on GitLab:

    git push origin --force 'refs/tags/*'

    Protected tags will cause this to fail. To proceed, you must remove tag protection, push, and then re-enable protected tags.

  6. To prevent dead links to commits that no longer exist, push the refs/replace created by git filter-repo.

    git push origin --force 'refs/replace/*'

    Refer to the Git replace documentation for information on how this works.

  7. Run a repository cleanup.

NOTE: Note: Project statistics are cached for performance. You may need to wait 5-10 minutes to see a reduction in storage utilization.

Purge files from GitLab storage

In addition to the refs mentioned above, GitLab also creates hidden refs/keep-around/*to prevent commits being deleted. Hidden refs are not advertised, which means we can't download them using Git, but these refs are included in a project export.

To purge files from GitLab storage:

  1. Install git filter-repo using a supported package manager or from source.

  2. Generate a fresh export from the project and download it.

  3. Decompress the backup using tar:

    tar xzf project-backup.tar.gz

    This will contain a project.bundle file, which was created by git bundle.

  4. Clone a fresh copy of the repository from the bundle:

    git clone --bare --mirror /path/to/project.bundle
  5. Using git filter-repo, purge any files from the history of your repository. Because we are trying to remove internal refs, we will rely on the commit-map produced by each run to tell us which internal refs to remove.

    NOTE: Note: git filter-repo creates a new commit-map file every run, and overwrite the commit-map from the previous run. You will need this file from every run. Do the next step every time you run git filter-repo.

    To purge all large files, the --strip-blobs-bigger-than option can be used:

    git filter-repo --strip-blobs-bigger-than 10M

    To purge specific large files by path, the --path and --invert-paths options can be combined.

    git filter-repo --path path/to/big/file.m4v --invert-paths

    See the git filter-repo documentation for more examples and the complete documentation.

  6. Because cloning from a bundle file sets the origin remote to the local bundle file, delete this origin remote, and set it to the URL to your repository:

    git remote remove origin
    git remote add origin https://gitlab.example.com/<namespace>/<project_name>.git
  7. Force push your changes to overwrite all branches on GitLab:

    git push origin --force 'refs/heads/*'

    Protected branches will cause this to fail. To proceed, you must remove branch protection, push, and then re-enable protected branches.

  8. To remove large files from tagged releases, force push your changes to all tags on GitLab:

    git push origin --force 'refs/tags/*'

    Protected tags will cause this to fail. To proceed, you must remove tag protection, push, and then re-enable protected tags.

  9. To prevent dead links to commits that no longer exist, push the refs/replace created by git filter-repo.

    git push origin --force 'refs/replace/*'

    Refer to the Git replace documentation for information on how this works.

  10. Run a repository cleanup.

Repository cleanup

NOTE: Note: Safely cleaning the repository requires it to be made read-only for the duration of the operation. This happens automatically, but submitting the cleanup request will fail if any writes are ongoing, so cancel any outstanding git push operations before continuing.

Introduced in GitLab 11.6.

Repository cleanup allows you to upload a text file of objects and GitLab will remove internal Git references to these objects. You can use git filter-repo to produce a list of objects (in a commit-map file) that can be used with repository cleanup.

To clean up a repository:

  1. Go to the project for the repository.

  2. Navigate to Settings > Repository.

  3. Upload a list of objects. For example, a commit-map file created by git filter-repo which is located in the filter-repo directory.

    If your commit-map file is larger than 10MB, the file can be split and uploaded piece by piece:

    split -l 100000 filter-repo/commit-map filter-repo/commit-map-
  4. Click Start cleanup.

This will:

  • Remove any internal Git references to old commits.
  • Run git gc against the repository to remove unreferenced objects. Repacking your repository will temporarily cause the size of your repository to increase significantly, because the old pack files are not removed until the new pack files have been created.
  • Unlink any unused LFS objects currently attached to your project, freeing up storage space.
  • Recalculate the size of your repository on disk.

You will receive an email notification with the recalculated repository size after the cleanup has completed.

When using repository cleanup, note:

  • Project statistics are cached. You may need to wait 5-10 minutes to see a reduction in storage utilization.
  • Housekeeping prunes loose objects older than 2 weeks. This means objects added in the last 2 weeks will not be removed immediately. If you have access to the Gitaly server, you may run git gc --prune=now to prune all loose objects immediately.
  • This process will remove some copies of the rewritten commits from GitLab's cache and database, but there are still numerous gaps in coverage and some of the copies may persist indefinitely. Clearing the instance cache may help to remove some of them, but it should not be depended on for security purposes!

Storage limits

Repository size limits:

When a project has reached its size limit, you cannot:

  • Push to the project.
  • Create a new merge request.
  • Merge existing merge requests.
  • Upload LFS objects.

You can still:

  • Create new issues.
  • Clone the project.

If you exceed the repository size limit, you can:

  1. Remove some data.
  2. Make a new commit.
  3. Push back to the repository.

If these actions are insufficient, you can also:

  • Move some blobs to LFS.
  • Remove some old dependency updates from history.

Unfortunately, this workflow doesn't work. Deleting files in a commit doesn't actually reduce the size of the repository, because the earlier commits and blobs still exist. Instead, you must rewrite history. We recommend the open-source community-maintained tool git filter-repo.

NOTE: Note: Until git gc runs on the GitLab side, the "removed" commits and blobs will still exist. You also must be able to push the rewritten history to GitLab, which may be impossible if you've already exceeded the maximum size limit.

In order to lift these restrictions, the administrator of the self-managed GitLab instance must increase the limit on the particular project that exceeded it. Therefore, it's always better to proactively stay underneath the limit. If you hit the limit, and can't have it temporarily increased, your only option is to:

  1. Prune all the unneeded stuff locally.
  2. Create a new project on GitLab and start using that instead.

CAUTION: Caution: This process is not suitable for removing sensitive data like password or keys from your repository. Information about commits, including file content, is cached in the database, and will remain visible even after they have been removed from the repository.

Troubleshooting

Incorrect repository statistics shown in the GUI

If the displayed size or commit number is different from the exported .tar.gz or local repository, you can ask a GitLab administrator to force an update.