Home

Awesome

Registry Cleaner

Stand With Ukraine cleanreg

This is a small tool to delete tags, or, to be more correct, delete image manifests, from a Docker Registry implementing the API v2.

Please be aware of that this is a soft delete. You've to run the registry garbage collection after this tool has been applied.

Information about the needed garbage collection is described at https://docs.docker.com/registry/garbage-collection/.

Credits and Contributors

The list of contributors can be found in the authors list in the python code.

This project uses shunit2 by Kate Ward.

History

Prerequisites and supported Platform

You need Docker and a Docker registry system which implements the Registry API v2. For Docker Registry v2 API specification see https://docs.docker.com/registry/spec/api/.

Be sure to configure your registry server to allow deletion (see https://docs.docker.com/registry/configuration/#/delete).

Usage

Download the file cleanreg.py or clone this repository to a local directory or pull the docker image. Beginning with version v0.8.0, images are supported for linux/amd64 and (this is new) linux\arm64. This is the suggested way to run cleanreg!

docker pull hcguersoy/cleanreg:v0.8.1

The image is hosted here: https://hub.docker.com/r/hcguersoy/cleanreg/

Hint: latest tag is not supported (you know, latest is evil :imp:).

To get a immediate help simply run it without any parameters:

$ docker run --rm hcguersoy/cleanreg:v0.8.1
usage: cleanreg.py [-h] [-v] -r REGISTRY [-p] [-y] [-q] [-n REPONAME:TAG]
                   [-k KEEPIMAGES] [-re] [-d DATE]
                   [-f REPOSFILE] [-c CACERT] [-sv] [-i] [-u BASICAUTHUSER]
                   [-pw BASICAUTHPW] [-w MD_WORKERS]

Removes images on a docker registry (v2).

optional arguments:
  -h, --help            show this help message and exit
  -v, --verbose         The verbosity level. Increase verbosity by multiple
                        usage, e.g. -vvv .
  -r REGISTRY, --registry REGISTRY
                        The registry server to connect to, e.g.
                        http://1.2.3.4:5000
  -p, --proxy           Use system level proxy settings accessing registry
                        server if set. By default, the registry server will be
                        accessed without a proxy.
  -y, --yes, --assume-yes
                        If set no user action will appear and all questions
                        will be answered with YES
  -q, --quiet           [deprecated] If set no user action will appear and all
                        questions will be answered with YES
  -n REPONAME, --reponame REPONAME:TAG
                        The name of the repo which should be cleaned up. Tags
                        are optional.
  -cf, --clean-full-catalog
                        If set all repos of the registry will be cleaned up,
                        considering the -k, -re and -d options.
                        These can be overridden for each repo in the repofile (-f).
  -k KEEPIMAGES, --keepimages KEEPIMAGES
                        Amount of images (not tags!) which should be kept for
                        the given repo  (if -n is set) or for each repo of the
                        registry (if -cf is set).
  -re, --regex
                        Interpret tagnames as regular expressions for the given
                        repo  (if -n is set) or for each repo of the registry (if
                        -cf is set).
  -s DATE, --since DATE
                        Keeps images which were created since then for
                        the given repo  (if -n is set) or for each repo of the
                        registry (if -cf is set).
                        Format: YYYYMMDD, YYYYMMDDThhmmss, YYYY-MM-DD or YYYY-MM-DDThh:mm:ss
  -f REPOSFILE, --reposfile REPOSFILE
                        A yaml file containing the list of Repositories with
                        additional information regarding tags, dates and how many
                        images to keep.
                        Format: REPONAME:
                            tag: TAG
                            keepimages: KEEPIMAGES
                            keepsince: DATE
  -c CACERT, --cacert CACERT
                        Path to a valid CA certificate file. This is needed if
                        self signed TLS is used in the registry server.
  -sv, --skip-tls-verify
                        If set insecure TLS is allowed, so no need for
                        a valid cert to verify.
  -i, --ignore-ref-tags
                        Ignore a digest if it is referenced multiple times in
                        the whole registry server. In this case, a list of all
                        repositories and their images will be retrieved which
                        can be time and memory consuming. ATTENTION: the
                        default is False so an image will be deleted even it
                        is referenced multiple times.
  -u BASICAUTHUSER, --basicauth-user BASICAUTHUSER
                        The username, if the registry is protected with basic
                        auth
  -pw BASICAUTHPW, --basicauth-pw BASICAUTHPW
                        The password, if the registry is protected with basic
                        auth
  -w MD_WORKERS, --metadata-workers MD_WORKERS
                        Parallel workers to retrieve image metadata. Default
                        value is 6.

If you've to use a configuration file (parameter -f) you should mount that file into your container:

docker run --rm -it -v $(pwd)/cleanreg-example.conf:/cleanreg-example.yaml hcguersoy/cleanreg:<version> -r  http://192.168.56.2:5000 -f cleanreg-example.yaml -i

As an alternative, you can create your own image, inheriting from cleanreg:

FROM hcguersoy/cleanreg:<version>
ADD myconfig.yaml /config.yaml

Examples

:exclamation: It is strongly recommended that you use the -i flag even it is more time and memory consuming. If not you can delete images / layers which you not wanted to delete because registry itself doesn't check if a digest is referenced by multiple tags!

Cleaning up a single repository called mysql on registry server 192.168.56.2:5000 and keeping 5 of the latest images:

docker run --rm -it hcguersoy/cleanreg:<version> -r http://192.168.56.2:5000 -n mysql -k 5

Be aware that you don't keep here the five last tags but digests/images. As a digest can be associated with multiple tags this can result in deletion of images which you not intended in! Again: to be secure use the -i flag:

docker run --rm -it hcguersoy/cleanreg:<version> -r http://192.168.56.2:5000 -n mysql -k 5 -i

Same as above but ignore images which are associated with multiple tags.

docker run --rm -it hcguersoy/cleanreg:<version> -r http://192.168.56.2:5000 -n mysql:latest -i

Will only delete the image mysql which is tagged as latest.

docker run --rm -it hcguersoy/cleanreg:<version> -r http://192.168.56.2:5000 -n mysql:.*temp.* -re -d 2022-01-01 -k 5 -i

Removes all images that contain the word "temp" in their tag's and if they were created before 2022 but at least 5 will be kept in total.

docker run --rm -it hcguersoy/cleanreg:<version> -r http://192.168.56.2:5000 -n myalpine -k 50 -i -w 12

If you have a very large registry and enough bandwidth you can increase the parallel workers to retrieve the image metadata. The default is 6. Be aware that you can generate a DoS on your registry server by increasing to much.

Cleaning up all repositories of the registry:

docker run --rm -it hcguersoy/cleanreg:<version> -r http://192.168.56.2:5000 -cf -k 5 -i

This will clean up all repositories, keeping 5 images per repository.

Cleaning up multiple repositories defined in a configuration file:

docker run --rm -it \
        -v $(pwd)/cleanreg-example.conf:/cleanreg-example.yaml \
       hcguersoy/cleanreg:<version>  \
       -r http://192.168.56.2:5000 \
       -f /cleanreg-example.yaml \
       -re \
       -i

The configuration file has the format

<repository name>:
    tag: <tag>
    keepimages: <number of images to keep>
    keepsince: <date>

The values for tag, keepimages and keepsince are optional. If the tag should be parsed as a regular expression use the -re flag as shown above. A simple example for the configuration file:

consul:
  tag: OnlyThisTag
  keepimages: 20
  keepsince: 2000-12-24T12:31:59
elasticsearch:
  keepimages: 20
  keepsince: 20000101
dummybox:
  keepimages: 0

The configuration file can be used together with the clean-full-catalog option:

./cleanreg.py -r http://192.168.56.2:5000 -cf -d 20180101 -f cleanreg-example.conf -i

This will clean the repositories with images to keep as defined in the configuration file and it will additionally clean all other repositories of the registry, keeping images per repository that were created since 2018.

There is a simple script added to create multiple image tags (based on busybox) on your registry server.

If you have installed a semi secure registry server using TLS and self signed certificates you have to provide the path to the CA certificate file:

./cleanreg.py -r https://192.168.56.3:5000 -c /my/certificates/ca.pem -f cleanreg-example.conf -i

If you run cleanreg in a container you should not forget to mount the certificate file into the container like the configuration file above. Alternatively you can set the option --skip-tls-verify. In this case, you don't need to provide a certificate.

:exclamation: be aware that this should only used for (local) testing and development but not for productive environments.

If your registry is protected with basic auth and the username is test and the password is secret, you have to pass these credentials to cleanreg.

./cleanreg.py -r https://192.168.56.3:5000 -u test -pw secret -f cleanreg-example.conf

:exclamation: if you provide the password this way, the password will be saved in your shell history in cleartext!

Running Garbage Collection

Example on running the garbage collection:

$ docker run --rm \
  -v /docker/registry2:/var/lib/registry:rw \
  registry:latest bin/registry \
  garbage-collect /etc/docker/registry/config.yml

This maps the local directory /docker/registry2 into the container, and calls the garbage collection. The pointed config file is the default configuration. The registry itself should be stopped before running this.

Contribution

Feel free to contribute your changes as a PR. Please ensure that the tests run without errors and provide tests for additional functionality.

It may take some time until I get aware of your PR and some time, too, until I merge it. Sorry for this.

This tool was initially implemented and tested on Ubuntu Linux 16.04 and on MacOS 10.13 using Python 2.7 and migrated to Python 3.10. It is developed and tested currently against Docker Registry version v2.8.1 on macOS 12 (arm64). The testing with older versions is dropped with version 0.8.1. It should work with older registry versions which are supporting the API, but they are not tested anymore.

You need to install the Python modules requests and PyYAML:

$ pip install requests PyYAML

Be sure to configure your registry server to allow deletion (see https://docs.docker.com/registry/configuration/#delete).

Run tests locally

Prerequisites:

You can run all tests, with the runAllTests.sh script:

cd test
./runAllTests.sh

This will run all tests and repeat them for different versions of the Docker Registry.

To run a single test, change to the test/tests directory and run a test script:

cd test/tests
./simple_clean.sh

By default the test will start the Docker Registry from Docker Hub with the tag latest. To specify another registry version, set it using the environment variable REGISTRYTAG.

cd test/tests
export REGISTRYTAG=2.5.1
./simple_clean.sh

GitHub Actions

After pushing to master or a branch feature\* or pr\* a GitHub Action will be triggered which runs the tests defined in test/runAllTests.sh. If one of these tests fail, the action will fail, too. If the build was triggered due to the creation of a release, a GitHub Action will be triggered, too. In this case, after running successfully the tests, the Docker image will be build with the help of buildx. Currently, two kinds of architectures are supported: linux/amd64 and linux/arm64. The later one could be used on Apple Silicon systems this the according runtime, e. g. Rancher Desktop.

Due to the lack of knowledge regarding other ARM systems (e. g. Raspberry Pi) you are very welcome to provide a PR.