Script for removing GitLab Job Artifacts.

remove_gitlab_artifacts.sh

#!/bin/bash
#
# Written by Chris Arceneaux
# GitHub: https://github.com/carceneaux
# Email: [email protected]
# Website: http://arsano.ninja
#
# Note: This code is a stop-gap to erase Job Artifacts for a project. I HIGHLY recommend you leverage
#       "artifacts:expire_in" in your .gitlab-ci.yml
#
# https://docs.gitlab.com/ee/ci/yaml/#artifactsexpire_in
#
# Software Requirements: curl, jq
#
# This code has been released under the terms of the Apache-2.0 license
# http://opensource.org/licenses/Apache-2.0


# project_id, find it here: https://gitlab.com/[organization name]/[repository name] at the top underneath repository name
project_id="207"

# token, find it here: https://gitlab.com/profile/personal_access_tokens
token="9hjGYpwmsMfBxT-Ghuu7"
server="gitlab.com"

# Retrieving Jobs list page count
total_pages=$(curl -sD - -o /dev/null -X GET \
  "https://$server/api/v4/projects/$project_id/jobs?per_page=100" \
  -H "PRIVATE-TOKEN: ${token}" | grep -Fi X-Total-Pages | sed 's/[^0-9]*//g')

# Creating list of Job IDs for the Project specified with Artifacts
job_ids=()
echo ""
echo "Creating list of all Jobs that currently have Artifacts..."
echo "Total Pages: ${total_pages}"
for ((i=2;i<=${total_pages};i++)) #starting with page 2 skipping most recent 100 Jobs
do
  echo "Processing Page: ${i}/${total_pages}"
  response=$(curl -s -X GET \
    "https://$server/api/v4/projects/$project_id/jobs?per_page=100&page=${i}" \
    -H "PRIVATE-TOKEN: ${token}")
  length=$(echo $response | jq '. | length')
  for ((j=0;j<${length};j++))
  do
    if [[ $(echo $response | jq ".[${j}].artifacts_file | length") > 0 ]]; then
        echo "Job found: $(echo $response | jq ".[${j}].id")"
        job_ids+=($(echo $response | jq ".[${j}].id"))
    fi
  done
done

# Loop through each Job erasing the Artifact(s)
echo ""
echo "${#job_ids[@]} Jobs found. Commencing removal of Artifacts..."
for job_id in ${job_ids[@]};
do
  response=$(curl -s -X DELETE \
    -H "PRIVATE-TOKEN:${token}" \
    "https://$server/api/v4/projects/$project_id/jobs/$job_id/artifacts")
  echo "Processing Job ID: ${job_id} - Status: $(echo $response | jq '.status')"
done

It is not removed artifacts from the 1st page of my pipelines list for some reason. It also missed .status attribute in console log. The rest is as advertised, thanks!

@Atarity Check the source code and you will find the hint #starting with page 2 skipping most recent 100 Jobs thus it is intended that the first page of artifacts are not removed.

The response can have a json with breaking lines \n. Consider removing it like ${response//\\n/}

response=${response//\\n/}
length=$(echo $response | jq '. | length')

Also, you can easy simulate next page checking [ $length -ne 0 ] and having the page loop to 1000 or more.

I made the following Python script, which works for over 10k jobs, too:

#!/usr/bin/env python3

import time

import requests

project_id = '...'
token = '...'
server = 'gitlab.com'

print("Creating list of all jobs that currently have artifacts...")
# We skip the first page.
url = f"https://{server}/api/v4/projects/{project_id}/jobs?per_page=100&page=2"
while url:
    print(f"Processing page: {url}")
    response = requests.get(
        url,
        headers={
            'private-token': token,
        },
    )

    if response.status_code in [500, 429]:
        print(f"Status {response.status_code}, retrying.")
        time.sleep(10)
        continue

    response.raise_for_status()
    response_json = response.json()
    for job in response_json:
        if job.get('artifacts_file', None):
            job_id = job['id']
            delete_response = requests.delete(
                f"https://{server}/api/v4/projects/{project_id}/jobs/{job_id}/artifacts",
                headers={
                    'private-token': token,
                },
            )
            print(f"Processing job ID: {job_id} - status: {delete_response.status_code}")

    url = response.links.get('next', {}).get('url', None)

The if job.get('artifacts_file', None): needs to be changed to if job.get('artifacts', None): in the current version of the API, at least I don't see artifacts_file in any of the JSON responses.

I see it here: https://docs.gitlab.com/ee/api/jobs.html

I don't know why but none of the jobs on our server had artifacts_file but artifacts instead where they were listed including their sizes etc.

"artifacts_file" worked for me, but it's trivial to support both, I also tweaked the output so you can see what job failed if any, and made it start at the first page:

#!/usr/bin/env python3

import time

import requests

project_id = '...'
token = '...'
server = 'gitlab.com'
start_page = 1

print("Creating list of all jobs that currently have artifacts...")
# We skip the first page.
url = f"https://{server}/api/v4/projects/{project_id}/jobs?per_page=100&page={start_page}"
while url:
    print(f"Processing page: {url}")
    response = requests.get(
        url,
        headers={
            'private-token': token,
        },
    )

    if response.status_code in [500, 429]:
        print(f"Status {response.status_code}, retrying.")
        time.sleep(10)
        continue

    response.raise_for_status()
    response_json = response.json()
    for job in response_json:
        if job.get('artifacts_file', None) or job.get('artifacts', None):
            job_id = job['id']
            print(f"Processing job ID: {job_id}", end="")
            delete_response = requests.delete(
                f"https://{server}/api/v4/projects/{project_id}/jobs/{job_id}/artifacts",
                headers={
                    'private-token': token,
                },
            )
            print(f" - status: {delete_response.status_code}")

    url = response.links.get('next', {}).get('url', None)

While the script deletes jobs' artifacts, you can also delete project's artifacts by adding this code:

url = f"https://{server}/api/v4/projects/{project_id}/artifacts"
delete_response = requests.delete(
    url,
    headers={
        'private-token': token,
    }
)
print(f" - status: {delete_response.status_code}")

This does not work if you're project has more than 10000 jobs, due to the removal of X-Total-Pages header from the Gitlab API responses.

Yes, I just found out that the X-Total-Pages header is now missing for performance reasons. Fortunately when a page number is too high, an empty json list ([]) is returned so it is quite easy to use a loop such as (here in bash):

PER_PAGE=100
PAGE=1
while JOBS=$(curl -s --header "PRIVATE-TOKEN: $TOKEN" "$GITLAB_INSTANCE/$PROJECT_ID/jobs?per_page=$PER_PAGE&page=$PAGE&sort=asc") && [ "$JOBS" != "[]" ]
do
   for JOB in $(echo $JOBS | jq .[].id)
   do
      [...]
   done
   PAGE=$((PAGE+1))
done

Here's my slightly improved version for the 'do it in python' section (ignores job.log files which seem to be non-deletable, uses command line arguments to load the settings):

#!/usr/bin/env python3

import time
import requests
import sys

server = sys.argv[1]
project_id = sys.argv[2]
token = sys.argv[3]
start_page = sys.argv[4]

print("Creating list of all jobs that currently have artifacts...")
# We skip the first page.
url = f"https://{server}/api/v4/projects/{project_id}/jobs?per_page=100&page={start_page}"
while url:
    print(f"Processing page: {url}")
    response = requests.get(
        url,
        headers={
            'private-token': token,
        },
    )

    if response.status_code in [500, 429]:
        print(f"Status {response.status_code}, retrying.")
        time.sleep(10)
        continue

    response.raise_for_status()
    response_json = response.json()
    for job in response_json:
        artifacts = job.get('artifacts_file', None)
        if not artifacts:
            artifacts = job.get('artifacts', None)

        has_artifacts = False
        for artifact in artifacts:
            if artifact['filename'] != 'job.log':
                has_artifacts = True
                break

        if has_artifacts:
            job_id = job['id']
            print(f"Processing job ID: {job_id}", end="")
            delete_response = requests.delete(
                f"https://{server}/api/v4/projects/{project_id}/jobs/{job_id}/artifacts",
                headers={
                    'private-token': token,
                },
            )
            print(f" - status: {delete_response.status_code}")

    url = response.links.get('next', {}).get('url', None)

Here's my slightly improved version for the 'do it in python' section (ignores job.log files which seem to be non-deletable, uses command line arguments to load the settings):

#!/usr/bin/env python3

import time
import requests
import sys

server = sys.argv[1]
project_id = sys.argv[2]
token = sys.argv[3]
start_page = sys.argv[4]

print("Creating list of all jobs that currently have artifacts...")
# We skip the first page.
url = f"https://{server}/api/v4/projects/{project_id}/jobs?per_page=100&page={start_page}"
while url:
    print(f"Processing page: {url}")
    response = requests.get(
        url,
        headers={
            'private-token': token,
        },
    )

    if response.status_code in [500, 429]:
        print(f"Status {response.status_code}, retrying.")
        time.sleep(10)
        continue

    response.raise_for_status()
    response_json = response.json()
    for job in response_json:
        artifacts = job.get('artifacts_file', None)
        if not artifacts:
            artifacts = job.get('artifacts', None)

        has_artifacts = False
        for artifact in artifacts:
            if artifact['filename'] != 'job.log':
                has_artifacts = True
                break

        if has_artifacts:
            job_id = job['id']
            print(f"Processing job ID: {job_id}", end="")
            delete_response = requests.delete(
                f"https://{server}/api/v4/projects/{project_id}/jobs/{job_id}/artifacts",
                headers={
                    'private-token': token,
                },
            )
            print(f" - status: {delete_response.status_code}")

    url = response.links.get('next', {}).get('url', None)

I get this error:

remove_artifacts.py", line 38, in <module>
    if artifact['filename'] != 'job.log':
       ~~~~~~~~^^^^^^^^^^^^
TypeError: string indices must be integers, not 'str'

@Tim-Schwalbe: Apologies, yes, I overlooked this case. I have amended the script to ignore artifacts_file, as this file seems to be contained in artifacts anyway.

I have improved my version a bit, it now automatically selects expired artifacts for deletion that (in my opinion) should be deleted in the first place, because they belong to jobs that were run on:

• merge requests that have been merged or closed;
• branches that have been merged.

It will also take a list of project ids as the last argument, making it easy to use in a cron job: Usage: {sys.argv[0]} <server> <token> <project id>...

#!/usr/bin/env python3

import time
import requests
import sys
from datetime import datetime, timezone
from dateutil import parser
import re

if len(sys.argv) < 4:
    print(f'Usage: {sys.argv[0]} <server> <token> <project id>...')

    exit(1)

server = sys.argv[1]
token = sys.argv[2]
project_ids = []
for i in range(3, len(sys.argv)):
    project_ids.append(sys.argv[i])


now = datetime.now(timezone.utc)

overall_space_savings = 0
for project_id in project_ids:
    print(f'Processing project {project_id}:')

    merge_request_url = f"https://{server}/api/v4/projects/{project_id}/merge_requests?scope=all&per_page=100&page=1"
    merge_requests = {}
    while merge_request_url:
        response = requests.get(
            merge_request_url,
            headers={
                'private-token': token,
            },
        )

        if response.status_code in [500, 429]:
            print(f"Status {response.status_code}, retrying.")
            time.sleep(10)
            continue

        response.raise_for_status()
        response_json = response.json()

        for merge_request in response_json:
            iid = merge_request.get('iid', None)
            if iid:
                merge_requests[int(iid)] = merge_request['state']

        merge_request_url = response.links.get('next', {}).get('url', None)

    branch_url = f"https://{server}/api/v4/projects/{project_id}/repository/branches?per_page=100&page=1"
    unmerged_branches = []
    while branch_url:
        response = requests.get(
            branch_url,
            headers={
                'private-token': token,
            },
        )

        if response.status_code in [500, 429]:
            print(f"Status {response.status_code}, retrying.")
            time.sleep(10)
            continue

        response.raise_for_status()
        response_json = response.json()

        for branch in response_json:
            is_merged = branch['merged']
            if not is_merged:
                unmerged_branches.append(branch['name'])

        branch_url = response.links.get('next', {}).get('url', None)


    url = f"https://{server}/api/v4/projects/{project_id}/jobs?per_page=100&page=1"

    job_count = 0
    artifact_count = 0
    artifact_size = 0
    deleted_artifact_count = 0
    deleted_artifact_size = 0
    while url:
        response = requests.get(
            url,
            headers={
                'private-token': token,
            },
        )

        if response.status_code in [500, 429]:
            print(f"Status {response.status_code}, retrying.")
            time.sleep(10)
            continue

        response.raise_for_status()
        response_json = response.json()
        for job in response_json:
            job_count += 1

            artifacts = job.get('artifacts', None)
            artifacts_expire_at_string = job.get('artifacts_expire_at', None)
            artifacts_expire_at = None
            if artifacts_expire_at_string:
                    artifacts_expire_at = parser.parse(artifacts_expire_at_string)

            has_expired_artifacts = False
            deleted_job_artifact_count = 0
            deleted_job_artifact_size = 0
            if artifacts:
                for artifact in artifacts:
                    if artifact['filename'] != 'job.log':
                        size = artifact['size']

                        artifact_count += 1
                        artifact_size += size

                        if not artifacts_expire_at or artifacts_expire_at < now:
                            has_expired_artifacts = True
                            deleted_job_artifact_count += 1
                            deleted_job_artifact_size += size


            delete_artifacts = False
            if has_expired_artifacts:
                ref = job['ref']
                merge_request_iid_match = re.search(r'refs\/merge-requests\/(\d+)\/head', ref)
                if merge_request_iid_match:
                    merge_request_iid = merge_request_iid_match.group(1)
                    if merge_request_iid:
                        merge_request_status = merge_requests.get(int(merge_request_iid))
                        if merge_request_status in ['merged', 'closed', None]:
                            delete_artifacts = True
                            deleted_artifact_count += deleted_job_artifact_count
                            deleted_artifact_size += deleted_job_artifact_size

                elif ref not in unmerged_branches:
                    delete_artifacts = True
                    deleted_artifact_count += deleted_job_artifact_count
                    deleted_artifact_size += deleted_job_artifact_size

            if delete_artifacts:
                job_id = job['id']
                print(f"Processing job ID: {job_id}", end="")
                delete_response = requests.delete(
                    f"https://{server}/api/v4/projects/{project_id}/jobs/{job_id}/artifacts",
                    headers={
                        'private-token': token,
                    },
                )
                print(f" - status: {delete_response.status_code}\033[K", end = "\r")


        print(f'Processed page {url}.\033[K', end = "\r")

        url = response.links.get('next', {}).get('url', None)

    overall_space_savings += deleted_artifact_size

    print()
    print(f'Jobs analysed: {job_count}');
    print(f'Pre artifact count: {artifact_count}');
    print(f'Pre artifact size [MB]: {artifact_size / (1024 * 1024)}')
    print(f'Post artifact count: {artifact_count - deleted_artifact_count}')
    print(f'Post artifact size [MB]: {(artifact_size - deleted_artifact_size) / (1024 * 1024)}')
    print()

print(f'Overall savings [MB]: {overall_space_savings / (1024 * 1024)}')

@mikeller I suggest you write your script code in a gist, or even fork this one here and replace it with you python code =)

Each gist indicates which forks have activity, making it easy to find interesting changes from others.

@voiski: Good point, done: https://gist.github.com/mikeller/ee7a668a83e4b9bc61646bddb4a2ade6

New version that takes a GitLab group id as a parameter and then cleans up all repositories in the group: https://gist.github.com/mikeller/7034d99bc27c361fc6a2df84e19c36ff