Reduce Container Registry Storage (FREE)

Container registries become large over time without cleanup. When a large number of images or tags are added:

  • Fetching the list of available tags or images becomes slower.
  • They take up a large amount of storage space on the server.

We recommend deleting unnecessary images and tags, and setting up a cleanup policy to automatically manage your container registry usage.

Check Container Registry Storage Use

The Usage Quotas page (Settings > Usage Quotas > Storage) displays storage usage for Packages. This page includes the Container Registry usage but is currently only available on GitLab.com. Measuring usage is only possible on the new version of the GitLab Container Registry backed by a metadata database. We are completing the upgrade and migration of the GitLab.com Container Registry first and only then will work on making this available to self-managed installs.

Image layers stored in the Container Registry are deduplicated at the root namespace level. Therefore, if you tag the same 500MB image twice (either in the same repository or across distinct repositories under the same root namespace), it will only count towards the root namespace usage once. Similarly, if a given image layer is shared across multiple images, be those under the same container repository, project, group, or across different ones, that layer will count only once towards the root namespace usage.

Only layers that are referenced by tagged images are accounted for. Untagged images and any layers referenced exclusively by them are subject to online garbage collection and automatically deleted after 24 hours if they remain unreferenced during that period.

Image layers are stored on the storage backend in the original (usually compressed) format. This means that the measured size for any given image layer should match the size displayed on the corresponding image manifest.

Cleanup policy

The cleanup policy is a scheduled job you can use to remove tags from the Container Registry. For the project where it's defined, tags matching the regex pattern are removed. The underlying layers and images remain.

To delete the underlying layers and images that aren't associated with any tags, administrators can use garbage collection with the -m switch.

Enable the cleanup policy

Cleanup policies can be run on all projects, with these exceptions:

  • For self-managed GitLab instances, the project must have been created in GitLab 12.8 or later. However, an administrator can enable the cleanup policy for all projects (even those created before 12.8) in GitLab application settings by setting container_expiration_policies_enable_historic_entries to true. Alternatively, you can execute the following command in the Rails console:

    ApplicationSetting.last.update(container_expiration_policies_enable_historic_entries: true)

    There are performance risks with enabling it for all projects, especially if you are using an external registry.

WARNING: For performance reasons, enabled cleanup policies are automatically disabled for projects on GitLab.com that don't have a container image.

How the cleanup policy works

The cleanup policy collects all tags in the Container Registry and excludes tags until only the tags to be deleted remain.

The cleanup policy searches for images based on the tag name. Support for the full path has not yet been implemented, but would allow you to clean up dynamically-named tags.

The cleanup policy:

  1. Collects all tags for a given repository in a list.
  2. Excludes the tag named latest from the list.
  3. Evaluates the name_regex (tags to expire), excluding non-matching names from the list.
  4. Excludes from the list any tags matching the name_regex_keep value (tags to preserve).
  5. Excludes any tags that do not have a manifest (not part of the options in the UI).
  6. Orders the remaining tags by created_date.
  7. Excludes from the list the N tags based on the keep_n value (Number of tags to retain).
  8. Excludes from the list the tags more recent than the older_than value (Expiration interval).
  9. Finally, the remaining tags in the list are deleted from the Container Registry.

WARNING: On GitLab.com, the execution time for the cleanup policy is limited, and some of the tags may remain in the Container Registry after the policy runs. The next time the policy runs, the remaining tags are included, so it may take multiple runs for all tags to be deleted.

WARNING: GitLab self-managed installations support third-party container registries that comply with the Docker Registry HTTP API V2 specification. However, this specification does not include a tag delete operation. Therefore, when interacting with third-party container registries, GitLab uses a workaround to delete tags. See the related issue for more information. Due to possible implementation variations, this workaround is not guaranteed to work with all third-party registries in the same predictable way. If you use the GitLab Container Registry, this workaround is not required because we implemented a special tag delete operation. In this case, you can expect cleanup policies to be consistent and predictable.

Create a cleanup policy

You can create a cleanup policy in the API or the UI.

To create a cleanup policy in the UI:

  1. For your project, go to Settings > Packages and registries.

  2. Expand the Clean up image tags section.

  3. Complete the fields.

    Field Description
    Toggle Turn the policy on or off.
    Run cleanup How often the policy should run.
    Keep the most recent How many tags to always keep for each image.
    Keep tags matching The regex pattern that determines which tags to preserve. The latest tag is always preserved. For all tags, use .*. See other regex pattern examples.
    Remove tags older than Remove only tags older than X days.
    Remove tags matching The regex pattern that determines which tags to remove. This value cannot be blank. For all tags, use .*. See other regex pattern examples.
  4. Select Save.

Depending on the interval you chose, the policy is scheduled to run.

NOTE: If you edit the policy and select Save again, the interval is reset.

Regex pattern examples

Cleanup policies use regex patterns to determine which tags should be preserved or removed, both in the UI and the API.

Regex patterns are automatically surrounded with \A and \Z anchors. Do not include any \A, \Z, ^ or $ token in the regex patterns as they are not necessary.

Here are examples of regex patterns you may want to use:

  • Match all tags:

    .*

    This is the default value for the expiration regex.

  • Match tags that start with v:

    v.+
  • Match only the tag named main:

    main
  • Match tags that are either named or start with release:

    release.*
  • Match tags that either start with v, are named main, or begin with release:

    (?:v.+|main|release.*)

Set cleanup limits to conserve resources

  • Introduced in GitLab 13.9 with a flag named container_registry_expiration_policies_throttling. Disabled by default.
  • Enabled by default in GitLab 14.9.
  • Removed the feature flag container_registry_expiration_policies_throttling in GitLab 15.0.

Cleanup policies are executed as a background process. This process is complex, and depending on the number of tags to delete, the process can take time to finish.

To prevent server resource starvation, the following application settings are available:

  • container_registry_expiration_policies_worker_capacity: the maximum number of cleanup workers running concurrently. This must be greater than or equal to 0. We recommend starting with a low number and increasing it after monitoring the resources used by the background workers. To remove all workers and not execute the cleanup policies, set this to 0. The default value is 4.
  • container_registry_delete_tags_service_timeout: the maximum time (in seconds) that the cleanup process can take to delete a batch of tags. The default value is 250.
  • container_registry_cleanup_tags_service_max_list_size: the maximum number of tags that can be deleted in a single execution. Additional tags must be deleted in another execution. We recommend starting with a low number and increasing it after monitoring that container images are properly deleted. The default value is 200.
  • container_registry_expiration_policies_caching: enable or disable tag creation timestamp caching during execution of policies. Cached timestamps are stored in Redis. Enabled by default.

For self-managed instances, those settings can be updated in the Rails console:

ApplicationSetting.last.update(container_registry_expiration_policies_worker_capacity: 3)

They are also available in the administrator area:

  1. On the top bar, select Main menu > Admin.
  2. Go to Settings > CI/CD > Container Registry.

Use the cleanup policy API

You can set, update, and disable the cleanup policies using the GitLab API.

Examples:

  • Select all tags, keep at least 1 tag per image, clean up any tag older than 14 days, run once a month, preserve any images with the name main and the policy is enabled:

    curl --request PUT --header 'Content-Type: application/json;charset=UTF-8' --header "PRIVATE-TOKEN: <your_access_token>" \
         --data-binary '{"container_expiration_policy_attributes":{"cadence":"1month","enabled":true,"keep_n":1,"older_than":"14d","name_regex":".*","name_regex_keep":".*-main"}}' \
         "https://gitlab.example.com/api/v4/projects/2"

Valid values for cadence when using the API are:

  • 1d (every day)
  • 7d (every week)
  • 14d (every two weeks)
  • 1month (every month)
  • 3month (every quarter)

Valid values for keep_n (number of tags kept per image name) when using the API are:

  • 1
  • 5
  • 10
  • 25
  • 50
  • 100

Valid values for older_than (days until tags are automatically removed) when using the API are:

  • 7d
  • 14d
  • 30d
  • 90d

See the API documentation for further details: Edit project API.

Use with external container registries

When using an external container registry, running a cleanup policy on a project may have some performance risks. If a project runs a policy to remove thousands of tags the GitLab background jobs may get backed up or fail completely. It is recommended you only enable container cleanup policies for projects that were created before GitLab 12.8 if you are confident the number of tags being cleaned up is minimal.

More Container Registry storage reduction options

Here are some other options to reduce your project's use of Container Registry storage:

Troubleshooting cleanup policies

Something went wrong while updating the cleanup policy.

If you see this error message, check the regex patterns to ensure they are valid.

GitLab uses RE2 syntax for regular expressions in the cleanup policy. You can test them with the regex101 regex tester. View some common regex pattern examples.

The cleanup policy doesn't delete any tags

There can be different reasons behind this:

  • In GitLab 13.6 and earlier, when you run the cleanup policy you may expect it to delete tags and it does not. This occurs when the cleanup policy is saved without editing the value in the Remove tags matching field. This field has a grayed out .* value as a placeholder. Unless .* (or another regex pattern) is entered explicitly into the field, a nil value is submitted. This value prevents the saved cleanup policy from matching any tags. As a workaround, edit the cleanup policy. In the Remove tags matching field, enter .* and save. This value indicates that all tags should be removed.

  • If you are on GitLab self-managed instances and you have 1000+ tags in a container repository, you might run into a Container Registry token expiration issue, with error authorizing context: invalid token in the logs.

    To fix this, there are two workarounds:

    • If you are on GitLab 13.9 or later, you can set limits for the cleanup policy. This limits the cleanup execution in time, and avoids the expired token error.

    • Extend the expiration delay of the Container Registry authentication tokens. This defaults to 5 minutes. You can set a custom value by running ApplicationSetting.last.update(container_registry_token_expire_delay: <integer>) in the Rails console, where <integer> is the desired number of minutes. For reference, 15 minutes is the value currently in use for GitLab.com. Be aware that by extending this value you increase the time required to revoke permissions.

If the previous fixes didn't work or you are on earlier versions of GitLab, you can generate a list of the tags that you want to delete, and then use that list to delete the tags. To do this, follow these steps:

  1. Run the following shell script. The command just before the for loop ensures that list_o_tags.out is always reinitialized when starting the loop. After running this command, all the tags' names will be in the list_o_tags.out file:

    # Get a list of all tags in a certain container repository while considering [pagination](../../../api/index.md#pagination)
    echo -n "" > list_o_tags.out; for i in {1..N}; do curl --header 'PRIVATE-TOKEN: <PAT>' "https://gitlab.example.com/api/v4/projects/<Project_id>/registry/repositories/<container_repo_id>/tags?per_page=100&page=${i}" | jq '.[].name' | sed 's:^.\(.*\).$:\1:' >> list_o_tags.out; done

    If you have Rails console access, you can enter the following commands to retrieve a list of tags limited by date:

    output = File.open( "/tmp/list_o_tags.out","w" )
    Project.find(<Project_id>).container_repositories.find(<container_repo_id>).tags.each do |tag|
      output << tag.name + "\n" if tag.created_at < 1.month.ago
    end;nil
    output.close

    This set of commands creates a /tmp/list_o_tags.out file listing all tags with a created_at date of older than one month.

  2. Remove from the list_o_tags.out file any tags that you want to keep. Here are some example sed commands for this. Note that these commands are simply examples. You may change them to best suit your needs:

    # Remove the `latest` tag from the file
    sed -i '/latest/d' list_o_tags.out
    
    # Remove the first N tags from the file
    sed -i '1,Nd' list_o_tags.out
    
    # Remove the tags starting with `Av` from the file
    sed -i '/^Av/d' list_o_tags.out
    
    # Remove the tags ending with `_v3` from the file
    sed -i '/_v3$/d' list_o_tags.out

    If you are running macOS, you must add .bak to the commands. For example:

    sed -i .bak '/latest/d' list_o_tags.out
  3. Double-check the list_o_tags.out file to make sure it contains only the tags that you want to delete.

  4. Run this shell script to delete the tags in the list_o_tags.out file:

    # loop over list_o_tags.out to delete a single tag at a time
    while read -r LINE || [[ -n $LINE ]]; do echo ${LINE}; curl --request DELETE --header 'PRIVATE-TOKEN: <PAT>' "https://gitlab.example.com/api/v4/projects/<Project_id>/registry/repositories/<container_repo_id>/tags/${LINE}"; sleep 0.1; echo; done < list_o_tags.out > delete.logs