Manual garbage collector for gitlab registry, it removes old revisions that are not referenced by any tag

gitlab-gc.sh

#!/bin/bash

BASE_PATH=/var/opt/gitlab/gitlab-rails/shared/registry/docker/registry/v2/repositories

DRY_RUN=0
KEEP_LAST_IMAGES=10
RUN_GARBAGE_COLLECTOR=0
GITLAB_CTL_COMMAND=`which gitlab-ctl`


if [ ! -x "${GITLAB_CTL_COMMAND}" ]; then
    echo "Missing gitlab-ctl command"
    exit 1
fi


while (( "$#" )); do
    case "$1" in

        "-b" | "--base-path")
            BASE_PATH=$2
            shift
            ;;

        "-r" | "--run-gc")
            RUN_GARBAGE_COLLECTOR=1
            ;;

        "-d"|"--dry-run")
            DRY_RUN=1
            ;;

        "-k"|"--keep")
            if ! ( echo $2 | grep -q '^[0-9]\+$') || [ $2 -eq 0 ]; then
                echo "Invalid value for keep last images '$2'"
                exit 1
            fi
            KEEP_LAST_IMAGES=$2
            shift
            ;;

        "-h"|"--help")
            echo "Usage: ${0} [options]"
            echo "Options:"
            echo -e "\t-k NUM, --keep NUM"
            echo -e "\t\tKeeps last NUM revisions, except current tags"
            echo
            echo -e "\t-d, --dry-run"
            echo -e "\t\tEnables dry run, no changes will be made"
            echo
            echo -e "\t-b, --base-path"
            echo -e "\t\tSets base path of Gitlab Registry repository storage"
            echo
            echo -e "\t-r, --run-gc"
            echo -e "\t\tStarts garbage collector after revision removal"
            exit 0
            ;;

        *)
            echo "Unknown argument: $1"
            exit 1
            ;;
    esac
    shift
done

IFS=$'\n'
used_hashes=`mktemp`
marked_hashes=`mktemp`
for repository in `find ${BASE_PATH} -mindepth 2 -maxdepth 2 -type d | sed "s#${BASE_PATH}/##"`; do
    for tag_hash in ${BASE_PATH}/${repository}/_manifests/tags/*/current/link; do 
        cat "${tag_hash}" | cut -d':' -f2; 
    done > "${used_hashes}"
    
    echo "Removing revisions of $repository:"
    ls -t ${BASE_PATH}/${repository}/_manifests/revisions/sha256 | fgrep -vf "${used_hashes}" | tail -n+${KEEP_LAST_IMAGES} | tee ${marked_hashes}
    if [ ${DRY_RUN} -ne 1 ]; then
        cat ${marked_hashes} | sed "s#^#${BASE_PATH}/${repository}/_manifests/revisions/sha256/#" | xargs rm -rf 
    fi
done
rm ${used_hashes}
rm ${marked_hashes}


if [ ${DRY_RUN} -eq 0 -a ${RUN_GARBAGE_COLLECTOR} -eq 1 ]; then
    "${GITLAB_CTL_COMMAND}" registry-garbage-collect
fi

Works like a charm in GitLab 9.1 EE. Thanks!

Thanks very very much for sharing!
I have created my own version removing the fixed maxdepth/mindepth of 2 and also including some extra control for directories that do not have valid "tags" or "revisions/sha256" directories, to avoid processing them.
Feel free to merge it to your release if you want, as the payload is exactly the same.
Regards!

This is my revision if you want to take a look:
https://gist.github.com/eedugon/69da0397c142e56f4186ee691502e7ee

Small features added:

• Searching in all BASE_PATH, not fixing the search to a depth of 2
• Directories without valid tags or revisions directories won't be processed (to avoid unexpected issues)
• Logging in case there's nothing to delete
• running registry-garbage-collect only if something was deleted
• Added crontab example