API requirements
Storage analysis
- Identify storage types
- Analyze storage in projects and groups
Manage CI/CD pipeline storage
Manage Container Registries storage
Manage package registry storage
Improve output readability
Testing for storage management automation
Community resources

Automate storage management

This page describes how to automate storage analysis and cleanup to manage your storage usage with the GitLab REST API.

You can also manage your storage usage by improving pipeline efficiency.

For more help with API automation, you can also use the GitLab community forum and Discord.

The script examples in this page are for demonstration purposes only and should not be used in production. You can use the examples to design and test your own scripts for storage automation.

API requirements

To automate storage management, your GitLab.com SaaS or self-managed instance must have access to the GitLab REST API.

API authentication scope

Use the following scopes to authenticate with the API:

Storage analysis:
- Read API access with the read_api scope.
- At least the Developer role on all projects.
Storage clean up:
- Full API access with the api scope.
- At least the Maintainer role on all projects.

You can use command-line tools or a programming language to interact with the REST API.

Command line tools

To send API requests, install either:

curl with your preferred package manager.
GitLab CLI and use the glab api subcommand.

To format JSON responses, install jq. For more information, see Tips for productive DevOps workflows: JSON formatting with jq and CI/CD linting automation.

To use these tools with the REST API:

curl

export GITLAB_TOKEN=xxx

curl --silent --header "Authorization: Bearer $GITLAB_TOKEN" "https://gitlab.com/api/v4/user" | jq

GitLab CLI

glab auth login

glab api groups/YOURGROUPNAME/projects

Using the GitLab CLI

Some API endpoints require pagination and subsequent page fetches to retrieve all results. The GitLab CLI provides the flag --paginate.

Requests that require a POST body formatted as JSON data can be written as key=value pairs passed to the --raw-field parameter.

For more information, see the GitLab CLI endpoint documentation.

API client libraries

The storage management and cleanup automation methods described in this page use:

The python-gitlab library, which provides a feature-rich programming interface.
The get_all_projects_top_level_namespace_storage_analysis_cleanup_example.py script in the GitLab API with Python project.

For more information about use cases for the python-gitlab library, see Efficient DevSecOps workflows: Hands-on python-gitlab API automation.

For more information about other API client libraries, see Third-party clients.

Use GitLab Duo Code Suggestions to write code more efficiently.

Storage analysis

Identify storage types

The projects API endpoint provides statistics for projects in your GitLab instance. To use the projects API endpoint, set the statistics key to boolean true. This data provides insight into storage consumption of the project by the following storage types:

storage_size: Overall storage
lfs_objects_size: LFS objects storage
job_artifacts_size: Job artifacts storage
packages_size: Packages storage
repository_size: Git repository storage
snippets_size: Snippets storage
uploads_size: Uploads storage
wiki_size: Wiki storage

To identify storage types:

curl

curl --silent --header "Authorization: Bearer $GITLAB_TOKEN" "https://gitlab.com/api/v4/projects/$GL_PROJECT_ID?statistics=true" | jq --compact-output '.id,.statistics' | jq
48349590
{
  "commit_count": 2,
  "storage_size": 90241770,
  "repository_size": 3521,
  "wiki_size": 0,
  "lfs_objects_size": 0,
  "job_artifacts_size": 90238249,
  "pipeline_artifacts_size": 0,
  "packages_size": 0,
  "snippets_size": 0,
  "uploads_size": 0
}

GitLab CLI

export GL_PROJECT_ID=48349590
glab api --method GET projects/$GL_PROJECT_ID --field 'statistics=true' | jq --compact-output '.id,.statistics' | jq
48349590
{
  "commit_count": 2,
  "storage_size": 90241770,
  "repository_size": 3521,
  "wiki_size": 0,
  "lfs_objects_size": 0,
  "job_artifacts_size": 90238249,
  "pipeline_artifacts_size": 0,
  "packages_size": 0,
  "snippets_size": 0,
  "uploads_size": 0
}

Python

project_obj = gl.projects.get(project.id, statistics=True)

print("Project {n} statistics: {s}".format(n=project_obj.name_with_namespace, s=json.dump(project_obj.statistics, indent=4)))

To print statistics for the project to the terminal, export the GL_GROUP_ID environment variable and run the script:

export GL_TOKEN=xxx
export GL_GROUP_ID=56595735

pip3 install python-gitlab
python3 get_all_projects_top_level_namespace_storage_analysis_cleanup_example.py

Project Developer Evangelism and Technical Marketing at GitLab  / playground / Artifact generator group / Gen Job Artifacts 4 statistics: {
    "commit_count": 2,
    "storage_size": 90241770,
    "repository_size": 3521,
    "wiki_size": 0,
    "lfs_objects_size": 0,
    "job_artifacts_size": 90238249,
    "pipeline_artifacts_size": 0,
    "packages_size": 0,
    "snippets_size": 0,
    "uploads_size": 0
}

Analyze storage in projects and groups

You can automate analysis of multiple projects and groups. For example, you can start at the top namespace level, and recursively analyze all subgroups and projects. You can also analyze different storage types.

Here’s an example of an algorithm to analyze multiple subgroups and projects:

Fetch the top-level namespace ID. You can copy the ID value from the namespace/group overview.
Fetch all subgroups from the top-level group, and save the IDs in a list.
Loop over all groups and fetch all projects from each group and save the IDs in a list.
Identify the storage type to analyze, and collect the information from project attributes, like project statistics, and job artifacts.
Print an overview of all projects, grouped by group, and their storage information.

The shell approach with glab might be more suitable for smaller analyses. For larger analyses, you should use a script that uses the API client libraries. This type of script can improve readability, data storage, flow control, testing, and reusability.

To ensure the script doesn’t reach API rate limits, the following example code is not optimized for parallel API requests.

To implement this algorithm:

GitLab CLI

export GROUP_NAME="gitlab-de"

# Return sub group IDs
glab api groups/$GROUP_NAME/subgroups | jq --compact-output '.[]' | jq --compact-output '.id'
12034712
67218622
67162711
67640130
16058698
12034604

# Loop over all subgroups to get subgroups, until the result set is empty. Example group: 12034712
glab api groups/12034712/subgroups | jq --compact-output '.[]' | jq --compact-output '.id'
56595735
70677315
67218606
70812167

# Lowest group level
glab api groups/56595735/subgroups | jq --compact-output '.[]' | jq --compact-output '.id'
# empty result, return and continue with analysis

# Fetch projects from all collected groups. Example group: 56595735
glab api groups/56595735/projects | jq --compact-output '.[]' | jq --compact-output '.id'
48349590
48349263
38520467
38520405

# Fetch storage types from a project (ID 48349590): Job artifacts in the `artifacts` key
glab api projects/48349590/jobs | jq --compact-output '.[]' | jq --compact-output '.id, .artifacts'
4828297946
[{"file_type":"archive","size":52444993,"filename":"artifacts.zip","file_format":"zip"},{"file_type":"metadata","size":156,"filename":"metadata.gz","file_format":"gzip"},{"file_type":"trace","size":3140,"filename":"job.log","file_format":null}]
4828297945
[{"file_type":"archive","size":20978113,"filename":"artifacts.zip","file_format":"zip"},{"file_type":"metadata","size":157,"filename":"metadata.gz","file_format":"gzip"},{"file_type":"trace","size":3147,"filename":"job.log","file_format":null}]
4828297944
[{"file_type":"archive","size":10489153,"filename":"artifacts.zip","file_format":"zip"},{"file_type":"metadata","size":158,"filename":"metadata.gz","file_format":"gzip"},{"file_type":"trace","size":3146,"filename":"job.log","file_format":null}]
4828297943
[{"file_type":"archive","size":5244673,"filename":"artifacts.zip","file_format":"zip"},{"file_type":"metadata","size":157,"filename":"metadata.gz","file_format":"gzip"},{"file_type":"trace","size":3145,"filename":"job.log","file_format":null}]
4828297940
[{"file_type":"archive","size":1049089,"filename":"artifacts.zip","file_format":"zip"},{"file_type":"metadata","size":157,"filename":"metadata.gz","file_format":"gzip"},{"file_type":"trace","size":3140,"filename":"job.log","file_format":null}]

Python

#!/usr/bin/env python

import datetime
import gitlab
import os
import sys

GITLAB_SERVER = os.environ.get('GL_SERVER', 'https://gitlab.com')
GITLAB_TOKEN = os.environ.get('GL_TOKEN') # token requires developer permissions
PROJECT_ID = os.environ.get('GL_PROJECT_ID') #optional
GROUP_ID = os.environ.get('GL_GROUP_ID') #optional

if __name__ == "__main__":
    if not GITLAB_TOKEN:
        print("🤔 Please set the GL_TOKEN env variable.")
        sys.exit(1)

    gl = gitlab.Gitlab(GITLAB_SERVER, private_token=GITLAB_TOKEN, pagination="keyset", order_by="id", per_page=100)

    # Collect all projects, or prefer projects from a group id, or a project id
    projects = []

    # Direct project ID
    if PROJECT_ID:
        projects.append(gl.projects.get(PROJECT_ID))
    # Groups and projects inside
    elif GROUP_ID:
        group = gl.groups.get(GROUP_ID)

        for project in group.projects.list(include_subgroups=True, get_all=True):
            manageable_project = gl.projects.get(project.id , lazy=True)
            projects.append(manageable_project)

    for project in projects:
        jobs = project.jobs.list(pagination="keyset", order_by="id", per_page=100, iterator=True)
        for job in jobs:
            print("DEBUG: ID {i}: {a}".format(i=job.id, a=job.attributes['artifacts']))

The script outputs the project job artifacts in a JSON formatted list:

[
    {
        "file_type": "archive",
        "size": 1049089,
        "filename": "artifacts.zip",
        "file_format": "zip"
    },
    {
        "file_type": "metadata",
        "size": 157,
        "filename": "metadata.gz",
        "file_format": "gzip"
    },
    {
        "file_type": "trace",
        "size": 3146,
        "filename": "job.log",
        "file_format": null
    }
]

Manage CI/CD pipeline storage

Job artifacts consume most of the pipeline storage, and job logs can also generate several hundreds of kilobytes. You should delete the unnecessary job artifacts first and then clean up job logs after analysis.

Deleting job log and artifacts is a destructive action that cannot be reverted. Use with caution. Deleting certain files, including report artifacts, job logs, and metadata files, affects GitLab features that use these files as data sources.

List job artifacts

To analyze pipeline storage, you can use the Job API endpoint to retrieve a list of job artifacts. The endpoint returns the job artifacts file_type key in the artifacts attribute. The file_type key indicates the artifact type:

archive is used for the generated job artifacts as a zip file.
metadata is used for additional metadata in a Gzip file.
trace is used for the job.log as a raw file.

Job artifacts provide a data structure that can be written as a cache file to disk, which you can use to test the implementation.

Based on the example code for fetching all projects, you can extend the Python script to do more analysis.

The following example shows a response from a query for job artifacts in a project:

[
    {
        "file_type": "archive",
        "size": 1049089,
        "filename": "artifacts.zip",
        "file_format": "zip"
    },
    {
        "file_type": "metadata",
        "size": 157,
        "filename": "metadata.gz",
        "file_format": "gzip"
    },
    {
        "file_type": "trace",
        "size": 3146,
        "filename": "job.log",
        "file_format": null
    }
]

Based on how you implement the script, you could either:

Collect all job artifacts and print a summary table at the end of the script.
Print the information immediately.

In the following example, job artifacts are collected in the ci_job_artifacts list. The script loops over all projects, and fetches:

The project_obj object variable that contains all attributes.
The artifacts attribute from the job object.

You can use keyset pagination to iterate over large lists of pipelines and jobs.

   ci_job_artifacts = []

    for project in projects:
        project_obj = gl.projects.get(project.id)

        jobs = project.jobs.list(pagination="keyset", order_by="id", per_page=100, iterator=True)

        for job in jobs:
            artifacts = job.attributes['artifacts']
            #print("DEBUG: ID {i}: {a}".format(i=job.id, a=json.dumps(artifacts, indent=4)))
            if not artifacts:
                continue

            for a in artifacts:
                data = {
                    "project_id": project_obj.id,
                    "project_web_url": project_obj.name,
                    "project_path_with_namespace": project_obj.path_with_namespace,
                    "job_id": job.id,
                    "artifact_filename": a['filename'],
                    "artifact_file_type": a['file_type'],
                    "artifact_size": a['size']
                }

                ci_job_artifacts.append(data)

    print("\nDone collecting data.")

    if len(ci_job_artifacts) > 0:
        print("|Project|Job|Artifact name|Artifact type|Artifact size|\n|-|-|-|-|-|") #Start markdown friendly table
        for artifact in ci_job_artifacts:
            print('| [{project_name}]({project_web_url}) | {job_name} | {artifact_name} | {artifact_type} | {artifact_size} |'.format(project_name=artifact['project_path_with_namespace'], project_web_url=artifact['project_web_url'], job_name=artifact['job_id'], artifact_name=artifact['artifact_filename'], artifact_type=artifact['artifact_file_type'], artifact_size=render_size_mb(artifact['artifact_size'])))
    else:
        print("No artifacts found.")

At the end of the script, job artifacts are printed as a Markdown formatted table. You can copy the table content to an issue comment or description, or populate a Markdown file in a GitLab repository.

$ python3 get_all_projects_top_level_namespace_storage_analysis_cleanup_example.py

|Project|Job|Artifact name|Artifact type|Artifact size|
|-|-|-|-|-|
| [gitlab-de/playground/artifact-gen-group/gen-job-artifacts-4](Gen Job Artifacts 4) | 4828297946 | artifacts.zip | archive | 50.0154 |
| [gitlab-de/playground/artifact-gen-group/gen-job-artifacts-4](Gen Job Artifacts 4) | 4828297946 | metadata.gz | metadata | 0.0001 |
| [gitlab-de/playground/artifact-gen-group/gen-job-artifacts-4](Gen Job Artifacts 4) | 4828297946 | job.log | trace | 0.0030 |
| [gitlab-de/playground/artifact-gen-group/gen-job-artifacts-4](Gen Job Artifacts 4) | 4828297945 | artifacts.zip | archive | 20.0063 |
| [gitlab-de/playground/artifact-gen-group/gen-job-artifacts-4](Gen Job Artifacts 4) | 4828297945 | metadata.gz | metadata | 0.0001 |
| [gitlab-de/playground/artifact-gen-group/gen-job-artifacts-4](Gen Job Artifacts 4) | 4828297945 | job.log | trace | 0.0030 |

Delete job artifacts in bulk

You can use a Python script to filter the types of job artifacts to delete in bulk.

Filter the API queries results to compare:

The created_at value to calculate the artifact age.
The size attribute to determine if artifacts meet the size threshold.

A typical request:

Deletes job artifacts older than the specified number of days.
Deletes job artifacts that exceed a specified amount of storage. For example, 100 MB.

In the following example, the script loops through job attributes and marks them for deletion. When the collection loops remove the object locks, the script deletes the job artifacts marked for deletion.

   for project in projects:
        project_obj = gl.projects.get(project.id)

        jobs = project.jobs.list(pagination="keyset", order_by="id", per_page=100, iterator=True)

        for job in jobs:
            artifacts = job.attributes['artifacts']
            if not artifacts:
                continue

            # Advanced filtering: Age and Size
            # Example: 90 days, 10 MB threshold (TODO: Make this configurable)
            threshold_age = 90 * 24 * 60 * 60
            threshold_size = 10 * 1024 * 1024

            # job age, need to parse API format: 2023-08-08T22:41:08.270Z
            created_at = datetime.datetime.strptime(job.created_at, '%Y-%m-%dT%H:%M:%S.%fZ')
            now = datetime.datetime.now()
            age = (now - created_at).total_seconds()
            # Shorter: Use a function
            # age = calculate_age(job.created_at)

            for a in artifacts:
                # ... removed analysis collection code for readability

                # Advanced filtering: match job artifacts age and size against thresholds
                if (float(age) > float(threshold_age)) or (float(a['size']) > float(threshold_size)):
                    # mark job for deletion (cannot delete inside the loop)
                    jobs_marked_delete_artifacts.append(job)

    print("\nDone collecting data.")

    # Advanced filtering: Delete all job artifacts marked to being deleted.
    for job in jobs_marked_delete_artifacts:
        # delete the artifact
        print("DEBUG", job)
        job.delete_artifacts()

    # Print collection summary (removed for readability)

Delete all job artifacts for a project

If you do not need the project’s job artifacts, you can use the following command to delete all job artifacts. This action cannot be reverted.

Artifact deletion can take several minutes or hours, depending on the number of artifacts to delete. Subsequent analysis queries against the API might return the artifacts as a false-positive result. To avoid confusion with results, do not immediately run additional API requests.

The artifacts for the most recent successful jobs are kept by default.

To delete all job artifacts for a project:

curl

export GL_PROJECT_ID=48349590

curl --silent --header "Authorization: Bearer $GITLAB_TOKEN" --request DELETE "https://gitlab.com/api/v4/projects/$GL_PROJECT_ID/artifacts"

GitLab CLI

glab api --method GET projects/$GL_PROJECT_ID/jobs | jq --compact-output '.[]' | jq --compact-output '.id, .artifacts'

glab api --method DELETE projects/$GL_PROJECT_ID/artifacts

Python

        project.artifacts.delete()

Delete job logs

When you delete a job log you also erase the entire job.

Example with the GitLab CLI:

glab api --method GET projects/$GL_PROJECT_ID/jobs | jq --compact-output '.[]' | jq --compact-output '.id'

4836226184
4836226183
4836226181
4836226180

glab api --method POST projects/$GL_PROJECT_ID/jobs/4836226180/erase | jq --compact-output '.name,.status'
"generate-package: [1]"
"success"

In the python-gitlab API library, use job.erase() instead of job.delete_artifacts(). To avoid this API call from being blocked, set the script to sleep for a short amount of time between calls that delete the job artifact:

    for job in jobs_marked_delete_artifacts:
        # delete the artifacts and job log
        print("DEBUG", job)
        #job.delete_artifacts()
        job.erase()
        # Sleep for 1 second
        time.sleep(1)

Support for creating a retention policy for job logs is proposed in issue 374717.

Delete old pipelines

Pipelines do not add to the overall storage usage, but if required you can automate their deletion.

To delete pipelines based on a specific date, specify the created_at key. You can use the date to calculate the difference between the current date and when the pipeline was created. If the age is larger than the threshold, the pipeline is deleted.

The created_at key must be converted from a timestamp to Unix epoch time, for example with date -d '2023-08-08T18:59:47.581Z' +%s.

Example with GitLab CLI:

export GL_PROJECT_ID=48349590

glab api --method GET projects/$GL_PROJECT_ID/pipelines | jq --compact-output '.[]' | jq --compact-output '.id,.created_at'
960031926
"2023-08-08T22:09:52.745Z"
959884072
"2023-08-08T18:59:47.581Z"

glab api --method DELETE projects/$GL_PROJECT_ID/pipelines/960031926

glab api --method GET projects/$GL_PROJECT_ID/pipelines | jq --compact-output '.[]' | jq --compact-output '.id,.created_at'
959884072
"2023-08-08T18:59:47.581Z"

In the following example that uses a Bash script:

jq and the GitLab CLI are installed and authorized.
The exported environment variable GL_PROJECT_ID.

The full script get_cicd_pipelines_compare_age_threshold_example.sh is located in the GitLab API with Linux Shell project.

#/bin/bash

CREATED_AT_ARR=$(glab api --method GET projects/$GL_PROJECT_ID/pipelines | jq --compact-output '.[]' | jq --compact-output '.created_at' | jq --raw-output @sh)

for row in ${CREATED_AT_ARR[@]}
do
    stripped=$(echo $row | xargs echo)
    #echo $stripped #DEBUG

    CREATED_AT_TS=$(date -d "$stripped" +%s)
    NOW=$(date +%s)

    AGE=$(($NOW-$CREATED_AT_TS))
    AGE_THRESHOLD=$((90*24*60*60)) # 90 days

    if [ $AGE -gt $AGE_THRESHOLD ];
    then
        echo "Pipeline age $AGE older than threshold $AGE_THRESHOLD, should be deleted."
        # TODO call glab to delete the pipeline. Needs an ID collected from the glab call above.
    else
        echo "Pipeline age $AGE not older than threshold $AGE_THRESHOLD. Ignore."
    fi
done

You can also use the python-gitlab API library and the created_at attribute to implement a similar algorithm that compares the job artifact age:

        # ...

        for pipeline in project.pipelines.list(iterator=True):
            pipeline_obj = project.pipelines.get(pipeline.id)
            print("DEBUG: {p}".format(p=json.dumps(pipeline_obj.attributes, indent=4)))

            created_at = datetime.datetime.strptime(pipeline.created_at, '%Y-%m-%dT%H:%M:%S.%fZ')
            now = datetime.datetime.now()
            age = (now - created_at).total_seconds()

            threshold_age = 90 * 24 * 60 * 60

            if (float(age) > float(threshold_age)):
                print("Deleting pipeline", pipeline.id)
                pipeline_obj.delete()

Automatic deletion of old pipelines is proposed in issue 338480.

List expiry settings for job artifacts

To manage artifact storage, you can update or configure when an artifact expires. The expiry setting for artifacts are configured in each job configuration in the .gitlab-ci.yml.

If there are multiple projects, and based on how job definitions are organized in the CI/CD configuration, it might be difficult to locate the expiry setting. You can use a script to search the entire CI/CD configuration. This includes access to objects that are resolved after they inherit values, like extends or !reference.

The script retrieves merged CI/CD configuration files and searches for the artifacts key to:

Identify jobs that do not have an expiry setting.
Return expiry settings for jobs that have the artifact expiry configured.

The following process describes how the script searches for the artifact expiry setting:

To generate a merged CI/CD configuration, the script loops over all projects and calls the ci_lint() method.
The yaml_load function loads the merged configuration into Python data structures for more analysis.
A dictionary that also has the key script identifies itself as a job definition, where the artifacts key might exists.
If yes, the script parses the sub key expire_in and stores the details to print later in a Markdown table summary.

    ci_job_artifacts_expiry = {}

    # Loop over projects, fetch .gitlab-ci.yml, run the linter to get the full translated config, and extract the `artifacts:` setting
    # https://python-gitlab.readthedocs.io/en/stable/gl_objects/ci_lint.html
    for project in projects:
            project_obj = gl.projects.get(project.id)
            project_name = project_obj.name
            project_web_url = project_obj.web_url
            try:
                lint_result = project.ci_lint.get()
                if lint_result.merged_yaml is None:
                    continue

                ci_pipeline = yaml.safe_load(lint_result.merged_yaml)
                #print("Project {p} Config\n{c}\n\n".format(p=project_name, c=json.dumps(ci_pipeline, indent=4)))

                for k in ci_pipeline:
                    v = ci_pipeline[k]
                    # This is a job object with `script` attribute
                    if isinstance(v, dict) and 'script' in v:
                        print(".", end="", flush=True) # Get some feedback that it is still looping
                        artifacts = v['artifacts'] if 'artifacts' in v else {}

                        print("Project {p} job {j} artifacts {a}".format(p=project_name, j=k, a=json.dumps(artifacts, indent=4)))

                        expire_in = None
                        if 'expire_in' in artifacts:
                            expire_in = artifacts['expire_in']

                        store_key = project_web_url + '_' + k
                        ci_job_artifacts_expiry[store_key] = { 'project_web_url': project_web_url,
                                                        'project_name': project_name,
                                                        'job_name': k,
                                                        'artifacts_expiry': expire_in}

            except Exception as e:
                 print(f"Exception searching artifacts on ci_pipelines: {e}".format(e=e))

    if len(ci_job_artifacts_expiry) > 0:
        print("|Project|Job|Artifact expiry|\n|-|-|-|") #Start markdown friendly table
        for k, details in ci_job_artifacts_expiry.items():
            if details['job_name'][0] == '.':
                continue # ignore job templates that start with a '.'
            print(f'| [{ details["project_name"] }]({details["project_web_url"]}) | { details["job_name"] } | { details["artifacts_expiry"] if details["artifacts_expiry"] is not None else "❌ N/A" } |')

The script generates a Markdown summary table with:

Project name and URL.
Job name.
The artifacts:expire_in setting, or N/A if there is no setting.

The script does not print job templates that:

Start with a . character.
Are not instantiated as runtime job objects that generate artifacts.

export GL_GROUP_ID=56595735

# Script requires pyyaml too.
pip3 install python-gitlab pyyaml

python3 get_all_cicd_config_artifacts_expiry.py

|Project|Job|Artifact expiry|
|-|-|-|
| [Gen Job Artifacts 4](https://gitlab.com/gitlab-de/playground/artifact-gen-group/gen-job-artifacts-4) | generator | 30 days |
| [Gen Job Artifacts with expiry and included jobs](https://gitlab.com/gitlab-de/playground/artifact-gen-group/gen-job-artifacts-expiry-included-jobs) | included-job10 | 10 days |
| [Gen Job Artifacts with expiry and included jobs](https://gitlab.com/gitlab-de/playground/artifact-gen-group/gen-job-artifacts-expiry-included-jobs) | included-job1 | 1 days |
| [Gen Job Artifacts with expiry and included jobs](https://gitlab.com/gitlab-de/playground/artifact-gen-group/gen-job-artifacts-expiry-included-jobs) | included-job30 | 30 days |
| [Gen Job Artifacts with expiry and included jobs](https://gitlab.com/gitlab-de/playground/artifact-gen-group/gen-job-artifacts-expiry-included-jobs) | generator | 30 days |
| [Gen Job Artifacts 2](https://gitlab.com/gitlab-de/playground/artifact-gen-group/gen-job-artifacts-2) | generator | ❌ N/A |
| [Gen Job Artifacts 1](https://gitlab.com/gitlab-de/playground/artifact-gen-group/gen-job-artifacts-1) | generator | ❌ N/A |

The get_all_cicd_config_artifacts_expiry.py script is located in the GitLab API with Python project.

Alternatively, you can use advanced search with API requests. The following example uses the scope: blobs to searches for the string artifacts in all *.yml files:

# https://gitlab.com/gitlab-de/playground/artifact-gen-group/gen-job-artifacts-expiry-included-jobs
export GL_PROJECT_ID=48349263

glab api --method GET projects/$GL_PROJECT_ID/search --field "scope=blobs" --field "search=expire_in filename:*.yml"

For more information about the inventory approach, see How GitLab can help mitigate deletion of open source container images on Docker Hub.

Set default expiry for job artifacts

To set the default expiry for job artifacts in a project, specify the expire_in value in the .gitlab-ci.yml file:

default:
    artifacts:
        expire_in: 1 week

Manage Container Registries storage

Container registries are available for projects or for groups. You can analyze both locations to implement a cleanup strategy.

List container registries

To list Container Registries in a project:

curl

export GL_PROJECT_ID=48057080

curl --silent --header "Authorization: Bearer $GITLAB_TOKEN" "https://gitlab.com/api/v4/projects/$GL_PROJECT_ID/registry/repositories" | jq --compact-output '.[]' | jq --compact-output '.id,.location' | jq
4435617
"registry.gitlab.com/gitlab-de/playground/container-package-gen-group/docker-alpine-generator"

curl --silent --header "Authorization: Bearer $GITLAB_TOKEN" "https://gitlab.com/api/v4/registry/repositories/4435617?size=true" | jq --compact-output '.id,.location,.size'
4435617
"registry.gitlab.com/gitlab-de/playground/container-package-gen-group/docker-alpine-generator"
3401613

GitLab CLI

export GL_PROJECT_ID=48057080

glab api --method GET projects/$GL_PROJECT_ID/registry/repositories | jq --compact-output '.[]' | jq --compact-output '.id,.location'
4435617
"registry.gitlab.com/gitlab-de/playground/container-package-gen-group/docker-alpine-generator"

glab api --method GET registry/repositories/4435617 --field='size=true' | jq --compact-output '.id,.location,.size'
4435617
"registry.gitlab.com/gitlab-de/playground/container-package-gen-group/docker-alpine-generator"
3401613

glab api --method GET projects/$GL_PROJECT_ID/registry/repositories/4435617/tags | jq --compact-output '.[]' | jq --compact-output '.name'
"latest"

glab api --method GET projects/$GL_PROJECT_ID/registry/repositories/4435617/tags/latest | jq --compact-output '.name,.created_at,.total_size'
"latest"
"2023-08-07T19:20:20.894+00:00"
3401613

Delete container images in bulk

When you delete container image tags in bulk, you can configure:

The matching regular expressions for tag names and images to keep (name_regex_keep) or delete (name_regex_delete)
The number of image tags to keep matching the tag name (keep_n)
The number of days before an image tag can be deleted (older_than)

On GitLab.com, due to the scale of the container registry, the number of tags deleted by this API is limited. If your container registry has a large number of tags to delete, only some of them are deleted. You might need to call the API multiple times. To schedule tags for automatic deletion, use a cleanup policy instead.

The following example uses the python-gitlab API library to fetch a list of tags, and calls the delete_in_bulk() method with filter parameters.

        repositories = project.repositories.list(iterator=True, size=True)
        if len(repositories) > 0:
            repository = repositories.pop()
            tags = repository.tags.list()

            # Cleanup: Keep only the latest tag
            repository.tags.delete_in_bulk(keep_n=1)
            # Cleanup: Delete all tags older than 1 month
            repository.tags.delete_in_bulk(older_than="1m")
            # Cleanup: Delete all tags matching the regex `v.*`, and keep the latest 2 tags
            repository.tags.delete_in_bulk(name_regex_delete="v.+", keep_n=2)

Create a cleanup policy for containers

Use the project REST API endpoint to create cleanup policies for containers. After you set the cleanup policy, all container images that match your specifications are deleted automatically. You do not need additional API automation scripts.

To send the attributes as a body parameter:

Use the --input - parameter to read from the standard input.
Set the Content-Type header.

The following example uses the GitLab CLI to create a cleanup policy:

export GL_PROJECT_ID=48057080

echo '{"container_expiration_policy_attributes":{"cadence":"1month","enabled":true,"keep_n":1,"older_than":"14d","name_regex":".*","name_regex_keep":".*-main"}}' | glab api --method PUT --header 'Content-Type: application/json;charset=UTF-8' projects/$GL_PROJECT_ID --input -

...

  "container_expiration_policy": {
    "cadence": "1month",
    "enabled": true,
    "keep_n": 1,
    "older_than": "14d",
    "name_regex": ".*",
    "name_regex_keep": ".*-main",
    "next_run_at": "2023-09-08T21:16:25.354Z"
  },

Optimize container images

You can optimize container images to reduce the image size and overall storage consumption in the container registry. Learn more in the pipeline efficiency documentation.

Manage package registry storage

Package registries are available for projects or for groups.

List packages and files

The following example shows fetching packages from a defined project ID using the GitLab CLI. The result set is an array of dictionary items that can be filtered with the jq command chain.

# https://gitlab.com/gitlab-de/playground/container-package-gen-group/generic-package-generator
export GL_PROJECT_ID=48377643

glab api --method GET projects/$GL_PROJECT_ID/packages | jq --compact-output '.[]' | jq --compact-output '.id,.name,.package_type'
16669383
"generator"
"generic"
16671352
"generator"
"generic"
16672235
"generator"
"generic"
16672237
"generator"
"generic"

Use the package ID to inspect the files and their size in the package.

glab api --method GET projects/$GL_PROJECT_ID/packages/16669383/package_files | jq --compact-output '.[]' |
 jq --compact-output '.package_id,.file_name,.size'

16669383
"nighly.tar.gz"
10487563

A similar automation shell script is created in the delete old pipelines section.

The following script example uses the python-gitlab library to fetch all packages in a loop, and loops over its package files to print the file_name and size attributes.

        packages = project.packages.list(order_by="created_at")

        for package in packages:

            package_files = package.package_files.list()
            for package_file in package_files:
                print("Package name: {p} File name: {f} Size {s}".format(
                    p=package.name, f=package_file.file_name, s=render_size_mb(package_file.size)))

Delete packages

Deleting a file in a package can corrupt the package. You should delete the package when performing automated cleanup maintenance.

To delete a package, use the GitLab CLI to change the --method parameter to DELETE:

glab api --method DELETE projects/$GL_PROJECT_ID/packages/16669383

To calculate the package size and compare it against a size threshold, you can use the python-gitlab library to extend the code described in the list packages and files section.

The following code example also calculates the package age and deletes the package when the conditions match:

        packages = project.packages.list(order_by="created_at")
        for package in packages:
            package_size = 0.0

            package_files = package.package_files.list()
            for package_file in package_files:
                print("Package name: {p} File name: {f} Size {s}".format(
                    p=package.name, f=package_file.file_name, s=render_size_mb(package_file.size)))

                package_size =+ package_file.size

            print("Package size: {s}\n\n".format(s=render_size_mb(package_size)))

            threshold_size = 10 * 1024 * 1024

            if (package_size > float(threshold_size)):
                print("Package size {s} > threshold {t}, deleting package.".format(
                    s=render_size_mb(package_size), t=render_size_mb(threshold_size)))
                package.delete()

            threshold_age = 90 * 24 * 60 * 60
            package_age = created_at = calculate_age(package.created_at)

            if (float(package_age > float(threshold_age))):
                print("Package age {a} > threshold {t}, deleting package.".format(
                    a=render_age_time(package_age), t=render_age_time(threshold_age)))
                package.delete()

The code generates the following output that you can use for further analysis:

Package name: generator File name: nighly.tar.gz Size 10.0017
Package size: 10.0017
Package size 10.0017 > threshold 10.0000, deleting package.

Package name: generator File name: 1-nightly.tar.gz Size 1.0004
Package size: 1.0004

Package name: generator File name: 10-nightly.tar.gz Size 10.0018
Package name: generator File name: 20-nightly.tar.gz Size 20.0033
Package size: 20.0033
Package size 20.0033 > threshold 10.0000, deleting package.

Dependency Proxy

Review the cleanup policy and how to purge the cache using the API

Improve output readability

You might need to convert timestamp seconds into a duration format, or print raw bytes in a more representative format. You can use the following helper functions to transform values for improved readability:

# Current Unix timestamp
date +%s

# Convert `created_at` date time with timezone to Unix timestamp
date -d '2023-08-08T18:59:47.581Z' +%s

Example with Python that uses the python-gitlab API library:

def render_size_mb(v):
    return "%.4f" % (v / 1024 / 1024)

def render_age_time(v):
    return str(datetime.timedelta(seconds = v))

# Convert `created_at` date time with timezone to Unix timestamp
def calculate_age(created_at_datetime):
    created_at_ts = datetime.datetime.strptime(created_at_datetime, '%Y-%m-%dT%H:%M:%S.%fZ')
    now = datetime.datetime.now()
    return (now - created_at_ts).total_seconds()

Testing for storage management automation

To test storage management automation, you might need to generate test data, or populate storage to verify that the analysis and deletion works as expected. The following sections provide tools and tips about testing and generating storage blobs in a short amount of time.

Generate job artifacts

Create a test project to generate fake artifact blobs using CI/CD job matrix builds. Add a CI/CD pipeline to generate artifacts on a daily basis

Create a new project.

Add the following snippet to .gitlab-ci.yml to include the job artifact generator configuration.

include:
    - remote: https://gitlab.com/gitlab-de/use-cases/efficiency/job-artifact-generator/-/raw/main/.gitlab-ci.yml

Configure pipeline schedules.
Trigger the pipeline manually.

Alternatively, reduce the 86 MB daily generated MB to different values in the MB_COUNT variable.

include:
    - remote: https://gitlab.com/gitlab-de/use-cases/efficiency/job-artifact-generator/-/raw/main/.gitlab-ci.yml

generator:
    parallel:
        matrix:
            - MB_COUNT: [1, 5, 10, 20, 50]

For more information, see the Job Artifact Generator README, with an example group.

Generate job artifacts with expiry

The project CI/CD configuration specifies job definitions in:

The main .gitlab-ci.yml configuration file.
The artifacts:expire_in setting.
Project files and templates.

To test the analysis scripts, the gen-job-artifacts-expiry-included-jobs project provides an example configuration.

# .gitlab-ci.yml
include:
    - include_jobs.yml

default:
  artifacts:
      paths:
          - '*.txt'

.gen-tmpl:
    script:
        - dd if=/dev/urandom of=${$MB_COUNT}.txt bs=1048576 count=${$MB_COUNT}

generator:
    extends: [.gen-tmpl]
    parallel:
        matrix:
            - MB_COUNT: [1, 5, 10, 20, 50]
    artifacts:
        untracked: false
        when: on_success
        expire_in: 30 days

# include_jobs.yml
.includeme:
    script:
        - dd if=/dev/urandom of=1.txt bs=1048576 count=1

included-job10:
    script:
        - echo "Servus"
        - !reference [.includeme, script]
    artifacts:
        untracked: false
        when: on_success
        expire_in: 10 days

included-job1:
    script:
        - echo "Gruezi"
        - !reference [.includeme, script]
    artifacts:
        untracked: false
        when: on_success
        expire_in: 1 days

included-job30:
    script:
        - echo "Grias di"
        - !reference [.includeme, script]
    artifacts:
        untracked: false
        when: on_success
        expire_in: 30 days

Generate container images

The example group container-package-gen-group provides projects that:

Use a base image in Dockerfile to build a new image.
Include the Docker.gitlab-ci.yml template to build images on GitLab.com SaaS.
Configure pipeline schedules to generate new images daily.

Example projects available to fork:

Generate generic packages

The example project generic-package-generator provides projects that:

Generate a random text blob, and create a tarball with the current Unix timestamp as release version.
Upload the tarball into the generic package registry, using the Unix timestamp as release version.

To generate generic packages, you can use this standalone .gitlab-ci.yml configuration:

generate-package:
  parallel:
    matrix:
      - MB_COUNT: [1, 5, 10, 20]
  before_script:
    - apt update && apt -y install curl
  script:
    - dd if=/dev/urandom of="${MB_COUNT}.txt" bs=1048576 count=${MB_COUNT}
    - tar czf "generated-$MB_COUNT-nighly-`date +%s`.tar.gz" "${MB_COUNT}.txt"
    - 'curl --header "JOB-TOKEN: $CI_JOB_TOKEN" --upload-file "generated-$MB_COUNT-nighly-`date +%s`.tar.gz" "${CI_API_V4_URL}/projects/${CI_PROJECT_ID}/packages/generic/generator/`date +%s`/${MB_COUNT}-nightly.tar.gz"'

  artifacts:
    paths:
      - '*.tar.gz'

Generate storage usage with forks

Use the following projects to test storage usage with cost factors for forks:

Fork gitlab-org/gitlab into a new namespace or group (includes LFS, Git repository).
Fork gitlab-com/www-gitlab-com into a new namespace or group.

Community resources

The following resources are not officially supported. Ensure to test scripts and tutorials before running destructive cleanup commands that may not be reverted.

Forum topic: Storage management automation resources
Script: GitLab Storage Analyzer, unofficial project by the GitLab Developer Evangelism team. You find similar code examples in this documentation how-to here.

Automate storage management

API requirements

API authentication scope

Command line tools

Using the GitLab CLI

API client libraries

Storage analysis

Identify storage types

Analyze storage in projects and groups

Manage CI/CD pipeline storage

List job artifacts

Delete job artifacts in bulk

Delete all job artifacts for a project

Delete job logs

Delete old pipelines

List expiry settings for job artifacts

Set default expiry for job artifacts

Manage Container Registries storage

List container registries

Delete container images in bulk

Create a cleanup policy for containers

Optimize container images

Manage package registry storage

List packages and files

Delete packages

Dependency Proxy

Improve output readability

Testing for storage management automation

Generate job artifacts

Generate job artifacts with expiry

Generate container images

Generate generic packages

Generate storage usage with forks

Community resources

Help & feedback

Docs

Product

Feature availability and product trials

Get Help