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.
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.
Purge files from repository history
To reduce the size of your repository in GitLab, you must first 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. -
refs/keep-around/*
are created as hidden refs to prevent commits referenced in the database from being removed
These refs are not automatically downloaded and hidden refs are not advertised, but we can remove these refs using a project export.
To purge files from a GitLab repository:
-
Install
git filter-repo
using a supported package manager or from source. -
Generate a fresh export from the project and download it. This project export contains a backup copy of your repository and refs we can use to purge files from your repository.
-
Decompress the backup using
tar
:tar xzf project-backup.tar.gz
This contains a
project.bundle
file, which was created bygit bundle
. -
Clone a fresh copy of the repository from the bundle using
--bare
and--mirror
options:git clone --bare --mirror /path/to/project.bundle
-
Using
git filter-repo
, purge any files from the history of your repository. Because we are trying to remove internal refs, we rely on thecommit-map
produced by each run to tell us which internal refs to remove.NOTE:
git filter-repo
creates a newcommit-map
file every run, and overwrite thecommit-map
from the previous run. You need this file from every run. Do the next step every time you rungit filter-repo
.To purge all files larger than 10M, 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. -
Because cloning from a bundle file sets the
origin
remote to the local bundle file, delete thisorigin
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
-
Force push your changes to overwrite all branches on GitLab:
git push origin --force 'refs/heads/*'
Protected branches cause this to fail. To proceed, you must remove branch protection, push, and then re-enable protected branches.
-
To remove large files from tagged releases, force push your changes to all tags on GitLab:
git push origin --force 'refs/tags/*'
Protected tags cause this to fail. To proceed, you must remove tag protection, push, and then re-enable protected tags.
-
To prevent dead links to commits that no longer exist, push the
refs/replace
created bygit filter-repo
.git push origin --force 'refs/replace/*'
Refer to the Git
replace
documentation for information on how this works. -
Run a repository cleanup.
Repository cleanup
Introduced in GitLab 11.6.
Repository cleanup allows you to upload a text file of objects and GitLab removes 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.
Introduced in GitLab 13.6,
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
fails if any writes are ongoing, so cancel any outstanding git push
operations before continuing.
To clean up a repository:
-
Go to the project for the repository.
-
Navigate to Settings > Repository.
-
Upload a list of objects. For example, a
commit-map
file created bygit filter-repo
which is located in thefilter-repo
directory.If your
commit-map
file is larger than about 250KB or 3000 lines, the file can be split and uploaded piece by piece:split -l 3000 filter-repo/commit-map filter-repo/commit-map-
-
Click Start cleanup.
This:
- Removes any internal Git references to old commits.
- Runs
git gc --prune=30.minutes.ago
against the repository to remove unreferenced objects. Repacking your repository temporarily causes the size of your repository to increase significantly, because the old pack files are not removed until the new pack files have been created. - Unlinks any unused LFS objects attached to your project, freeing up storage space.
- Recalculates the size of your repository on disk.
GitLab sends an email notification with the recalculated repository size after the cleanup has completed.
If the repository size does not decrease, this may be caused by loose objects being kept around because they were referenced in a Git operation that happened in the last 30 minutes. Try re-running these steps after the repository has been dormant for at least 30 minutes.
When using repository cleanup, note:
- Project statistics are cached. You may need to wait 5-10 minutes to see a reduction in storage utilization.
- The cleanup prunes loose objects older than 30 minutes. This means objects added or referenced in the last 30 minutes
are not be removed immediately. If you have access to the
Gitaly server, you may slip that delay and run
git gc --prune=now
to prune all loose objects immediately. - This process removes some copies of the rewritten commits from the GitLab 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:
- Can be set by an administrator on self-managed instances. (PREMIUM SELF)
- Are set for GitLab.com.
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:
- Remove some data.
- Make a new commit.
- 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:
Until git gc
runs on the GitLab side, the "removed" commits and blobs 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:
- Prune all the unneeded stuff locally.
- Create a new project on GitLab and start using that instead.
WARNING: 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 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.