GitLab Git Large File Storage (LFS) Administration

Version history
  • Git LFS is supported in GitLab starting with version 8.2.
  • Support for object storage, such as AWS S3, was introduced in 10.0.
  • LFS is enabled in GitLab self-managed instances by default.

Documentation about how to use Git LFS are under Managing large binary files with Git LFS doc.

Requirements

Configuration

Git LFS objects can be large in size. By default, they are stored on the server GitLab is installed on.

There are various configuration options to help GitLab server administrators:

  • Enabling/disabling Git LFS support.
  • Changing the location of LFS object storage.
  • Setting up object storage supported by Fog.

Configuration for Omnibus installations

In /etc/gitlab/gitlab.rb:

# Change to true to enable lfs - enabled by default if not defined
gitlab_rails['lfs_enabled'] = false

# Optionally, change the storage path location. Defaults to
# `#{gitlab_rails['shared_path']}/lfs-objects`. Which evaluates to
# `/var/opt/gitlab/gitlab-rails/shared/lfs-objects` by default.
gitlab_rails['lfs_storage_path'] = "/mnt/storage/lfs-objects"

After you update settings in /etc/gitlab/gitlab.rb, run Omnibus GitLab reconfigure.

Configuration for installations from source

In config/gitlab.yml:

# Change to true to enable lfs
  lfs:
    enabled: false
    storage_path: /mnt/storage/lfs-objects

Storing LFS objects in remote object storage

You can store LFS objects in remote object storage. This allows you to offload reads and writes to the local disk, and free up disk space significantly. GitLab is tightly integrated with Fog, so you can refer to its documentation to check which storage services can be integrated with GitLab. You can also use external object storage in a private local network. For example, MinIO is a standalone object storage service that works with GitLab instances.

GitLab provides two different options for the uploading mechanism: “Direct upload” and “Background upload”.

Read more about using object storage with GitLab.

noteIn GitLab 13.2 and later, we recommend using the consolidated object storage settings. This section describes the earlier configuration format.

Option 1. Direct upload

  1. User pushes an lfs file to the GitLab instance.
  2. GitLab-workhorse uploads the file directly to the external object storage.
  3. GitLab-workhorse notifies GitLab-rails that the upload process is complete.

Option 2. Background upload

  1. User pushes an lfs file to the GitLab instance.
  2. GitLab-rails stores the file in the local file storage.
  3. GitLab-rails then uploads the file to the external object storage asynchronously.

The following general settings are supported.

Setting Description Default
enabled Enable/disable object storage. false
remote_directory The bucket name where LFS objects are stored.  
direct_upload Set to true to enable direct upload of LFS without the need of local shared storage. Option may be removed after we decide to support only single storage for all files. false
background_upload Set to false to disable automatic upload. Option may be removed once upload is direct to S3. true
proxy_download Set to true to enable proxying all files served. Option allows to reduce egress traffic as this allows clients to download directly from remote storage instead of proxying all data. false
connection Various connection options described below.  

See the available connection settings for different providers.

Here is a configuration example with S3.

Manual uploading to an object storage

There are two ways to manually do the same thing as automatic uploading (described above).

Option 1: Rake task

gitlab-rake gitlab:lfs:migrate

Option 2: Rails console

Log into the Rails console:

sudo gitlab-rails console

Upload LFS files manually

LfsObject.where(file_store: [nil, 1]).find_each do |lfs_object|
  lfs_object.file.migrate!(ObjectStorage::Store::REMOTE) if lfs_object.file.file.exists?
end

S3 for Omnibus installations

On Omnibus GitLab installations, the settings are prefixed by lfs_object_store_:

  1. Edit /etc/gitlab/gitlab.rb and add the following lines, replacing values based on your needs:

    gitlab_rails['lfs_object_store_enabled'] = true
    gitlab_rails['lfs_object_store_remote_directory'] = "lfs-objects"
    gitlab_rails['lfs_object_store_connection'] = {
      'provider' => 'AWS',
      'region' => 'eu-central-1',
      'aws_access_key_id' => '1ABCD2EFGHI34JKLM567N',
      'aws_secret_access_key' => 'abcdefhijklmnopQRSTUVwxyz0123456789ABCDE',
      # The below options configure an S3 compatible host instead of AWS
      'host' => 'localhost',
      'endpoint' => 'http://127.0.0.1:9000',
      'path_style' => true
    }
    
  2. Save the file, and then reconfigure GitLab for the changes to take effect.
  3. Migrate any existing local LFS objects to the object storage:

    gitlab-rake gitlab:lfs:migrate
    

    This migrates existing LFS objects to object storage. New LFS objects are forwarded to object storage unless gitlab_rails['lfs_object_store_background_upload'] and gitlab_rails['lfs_object_store_direct_upload'] is set to false.

  4. (Optional) Verify all files migrated properly. From PostgreSQL console (sudo gitlab-psql -d gitlabhq_production) verify objectstg below (where file_store=2) has count of all artifacts:

    gitlabhq_production=# SELECT count(*) AS total, sum(case when file_store = '1' then 1 else 0 end) AS filesystem, sum(case when file_store = '2' then 1 else 0 end) AS objectstg FROM lfs_objects;
    
    total | filesystem | objectstg
    ------+------------+-----------
     2409 |          0 |      2409
    

    Verify no files on disk in artifacts folder:

    sudo find /var/opt/gitlab/gitlab-rails/shared/lfs-objects -type f | grep -v tmp/cache | wc -l
    

S3 for installations from source

For source installations the settings are nested under lfs: and then object_store::

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

    lfs:
    enabled: true
    object_store:
      enabled: false
      remote_directory: lfs-objects # Bucket name
      connection:
        provider: AWS
        aws_access_key_id: 1ABCD2EFGHI34JKLM567N
        aws_secret_access_key: abcdefhijklmnopQRSTUVwxyz0123456789ABCDE
        region: eu-central-1
        # Use the following options to configure an AWS compatible host such as Minio
        host: 'localhost'
        endpoint: 'http://127.0.0.1:9000'
        path_style: true
    
  2. Save the file, and then restart GitLab for the changes to take effect.
  3. Migrate any existing local LFS objects to the object storage:

    sudo -u git -H bundle exec rake gitlab:lfs:migrate RAILS_ENV=production
    

    This migrates existing LFS objects to object storage. New LFS objects are forwarded to object storage unless background_upload and direct_upload is set to false.

  4. (Optional) Verify all files migrated properly. From PostgreSQL console (sudo -u git -H psql -d gitlabhq_production) verify objectstg below (where file_store=2) has count of all artifacts:

    gitlabhq_production=# SELECT count(*) AS total, sum(case when file_store = '1' then 1 else 0 end) AS filesystem, sum(case when file_store = '2' then 1 else 0 end) AS objectstg FROM lfs_objects;
    
    total | filesystem | objectstg
    ------+------------+-----------
     2409 |          0 |      2409
    

    Verify no files on disk in artifacts folder:

    sudo find /var/opt/gitlab/gitlab-rails/shared/lfs-objects -type f | grep -v tmp/cache | wc -l
    

Migrating back to local storage

To migrate back to local storage:

  1. Set both direct_upload and background_upload to false under the LFS object storage settings. Don’t forget to restart GitLab.
  2. Run rake gitlab:lfs:migrate_to_local on your console.
  3. Disable object_storage for LFS objects in gitlab.rb. Remember to restart GitLab afterwards.

Storage statistics

You can see the total storage used for LFS objects on groups and projects:

Troubleshooting

Missing LFS objects

An error about a missing LFS object may occur in either of these situations:

  • When migrating LFS objects from disk to object storage, with error messages like:

    ERROR -- : Failed to transfer LFS object
    006622269c61b41bf14a22bbe0e43be3acf86a4a446afb4250c3794ea47541a7
    with error: No such file or directory @ rb_sysopen -
    /var/opt/gitlab/gitlab-rails/shared/lfs-objects/00/66/22269c61b41bf14a22bbe0e43be3acf86a4a446afb4250c3794ea47541a7
    

    (Line breaks have been added for legibility.)

  • When running the integrity check for LFS objects with the VERBOSE=1 parameter.

The database can have records for LFS objects which are not on disk. The database entry may prevent a new copy of the object from being pushed. To delete these references:

  1. Start a rails console.
  2. Query the object that’s reported as missing in the rails console, to return a file path:

    lfs_object = LfsObject.find_by(oid: '006622269c61b41bf14a22bbe0e43be3acf86a4a446afb4250c3794ea47541a7')
    lfs_object.file.path
    
  3. Check on disk or object storage if it exists:

    ls -al /var/opt/gitlab/gitlab-rails/shared/lfs-objects/00/66/22269c61b41bf14a22bbe0e43be3acf86a4a446afb4250c3794ea47541a7
    
  4. If the file is not present, remove the database record via the rails console:

    lfs_object.destroy
    

Google::Apis::TransmissionError: execution expired

If LFS integration is configured with Google Cloud Storage and background uploads (background_upload: true and direct_upload: false), Sidekiq workers may encounter this error. This is because the uploading timed out with very large files. LFS files up to 6 GB can be uploaded without any extra steps, otherwise you need to use the following workaround.

Sign in to Rails console:

sudo gitlab-rails console

Set up timeouts:

  • These settings are only in effect for the same session. For example, they are not effective for Sidekiq workers.
  • 20 minutes (1200 sec) is enough to upload 30GB LFS files:
::Google::Apis::ClientOptions.default.open_timeout_sec = 1200
::Google::Apis::ClientOptions.default.read_timeout_sec = 1200
::Google::Apis::ClientOptions.default.send_timeout_sec = 1200

Upload LFS files manually (this process does not use Sidekiq at all):

LfsObject.where(file_store: [nil, 1]).find_each do |lfs_object|
  lfs_object.file.migrate!(ObjectStorage::Store::REMOTE) if lfs_object.file.file.exists?
end

See more information in !19581

LFS commands fail on TLS v1.3 server

If you configure GitLab to disable TLS v1.2 and only enable TLS v1.3 connections, LFS operations require a Git LFS client version 2.11.0 or later. If you use a Git LFS client earlier than version 2.11.0, GitLab displays an error:

batch response: Post https://username:***@gitlab.example.com/tool/releases.git/info/lfs/objects/batch: remote error: tls: protocol version not supported
error: failed to fetch some objects from 'https://username:[MASKED]@gitlab.example.com/tool/releases.git/info/lfs'

When using GitLab CI over a TLS v1.3 configured GitLab server, you must upgrade to GitLab Runner 13.2.0 or later to receive an updated Git LFS client version via the included GitLab Runner Helper image.

To check an installed Git LFS client’s version, run this command:

git lfs version

Known limitations

  • Support for removing unreferenced LFS objects was added in 8.14 onward.
  • LFS authentications via SSH was added with GitLab 8.12.
  • Only compatible with the Git LFS client versions 1.1.0 and later, or 1.0.2.
  • The storage statistics count each LFS object for every project linking to it.