A simple Docker container and image garbage collection script.
- Containers that exited more than an hour ago are removed.
- Images that don't belong to any remaining container after that are removed.
- Optionally, remove volumes that are not associated to any remaining container after removal (Available only for docker >= 1.9.0)
Although docker normally prevents removal of images that are in use by
containers, we take extra care to not remove any image tags (e.g., ubuntu:14.04,
busybox, etc) that are in use by containers. A naive docker rmi $(docker images -q) will leave images stripped of all tags, forcing docker to re-pull the
repositories when starting new containers even though the images themselves are
still on disk.
This script is intended to be run as a cron job, but you can also run it as a Docker container (see below).
sudo apt-get install git devscripts debhelper build-essential dh-make
git clone https://github.com/spotify/docker-gc.git
cd docker-gc
debuild -us -uc -bIf you get lintian errors during debuild, try debuild --no-lintian -us -uc -b.
sudo dpkg -i ../docker-gc_0.1.0_all.debThis installs the docker-gc script into /usr/sbin. If you want it to
run as a cron job, you can configure it now by creating a root-owned
executable file /etc/cron.hourly/docker-gc with the following contents:
#!/bin/bash
/usr/sbin/docker-gc
To test that the job will actually run you can use this command
run-parts --test /etc/cron.hourly
To use the script manually, run docker-gc. The system user under
which docker-gc runs needs to have read and write access to
the $STATE_DIR environment variable which defaults to /var/lib/docker-gc.
There can be images that are large that serve as a common base for
many application containers, and as such, make sense to pin to the
machine, as many derivative containers will use it. This can save
time in pulling those kinds of images. There may be other reasons to
exclude images from garbage collection. To do so, create
/etc/docker-gc-exclude, or if you want the file to be read from
elsewhere, set the EXCLUDE_FROM_GC environment variable to its
location. This file can contain image name patterns (in the grep
sense), one per line, such as spotify/cassandra:latest or it can
contain image ids (truncated to the length shown in docker images
which is 12.
An example image excludes file might contain:
spotify/cassandra:latest
redis:.*
9681260c3ad5
There can also be containers (for example data only containers) which
you would like to exclude from garbage collection. To do so, create
/etc/docker-gc-exclude-containers, or if you want the file to be
read from elsewhere, set the EXCLUDE_CONTAINERS_FROM_GC environment
variable to its location. This file should contain name patterns (in
the grep sense), one per line, such as mariadb-data.
An example container excludes file might contain:
mariadb-data
drunk_goodall
There can be occasions where you don't want to remove a dangling volume.
To enable this functionality you can create a file named
/etc/docker-gc-exclude-volumes (or specify EXCLUDE_VOLUMES_IDS_FILE env var
with any path for such file), containing name patterns (in the grep sense),
one per line, of volumes that will be excluded from garbage collection.
By default, docker will not remove an image if it is tagged in multiple repositories. If you have a server running docker where this is the case, for example in CI environments where dockers are being built, re-tagged, and pushed, you can enable a force flag to override this default.
FORCE_IMAGE_REMOVAL=1 docker-gc
You might want to always keep a set of the most recent images for any
repository. For example, if you are continually rebuilding an image during
development you would want to clear out all but the most recent version of an
image. To do so, set the MINIMUM_IMAGES_TO_SAVE=1 environment variable. You
can preserve any count of the most recent images, e.g. save the most recent 10
with MINIMUM_IMAGES_TO_SAVE=10.
By default, if an error is encountered when cleaning up a container, Docker will report the error back and leave it on disk. This can sometimes lead to containers accumulating. If you run into this issue, you can force the removal of the container by setting the environment variable below:
FORCE_CONTAINER_REMOVAL=1 docker-gc
By default, docker-gc will not remove a container if it exited less than 3600 seconds (1 hour) ago. In some cases you might need to change this setting (e.g. you need exited containers to stick around for debugging for several days). Set the GRACE_PERIOD_SECONDS variable to override this default.
GRACE_PERIOD_SECONDS=86400 docker-gc
This setting also prevents the removal of images that have been created less than GRACE_PERIOD_SECONDS seconds ago.
By default, docker-gc will proceed with deletion of containers and images. To test your command-line options set the DRY_RUN variable to override this default.
DRY_RUN=1 docker-gc
A Dockerfile is provided as an alternative to a local installation. By default the container will start up, run a single garbage collection, and shut down.
The image is published as spotify/docker-gc.
The image is currently built with Docker 17.09.0-ce, but to build it against a newer
Docker version (to ensure that the API version of the command-line interface
matches with your Docker daemon), simply edit the ENV DOCKER_VERSION line in
Dockerfile prior to the build step below.
Build the Docker image with make -f Makefile.docker image or:
docker build -t spotify/docker-gc .The docker-gc container requires access to the docker socket in order to function, so you need to map it when running, e.g.:
docker run --rm --privileged -v /var/run/docker.sock:/var/run/docker.sock -v /etc:/etc:ro spotify/docker-gcThe /etc directory is also mapped so that it can read any exclude files
that you've created.
If you want to remove volumes, you can do so by passing REMOVE_VOLUMES env var set to 1.
$ docker run --rm -v /var/run/docker.sock:/var/run/docker.sock -v /etc:/etc -e REMOVE_VOLUMES=1 spotify/docker-gcIf you want to remove volumes only for a specified driver, you can do it by passing VOLUME_DELETE_ONLY_DRIVER env var set to the driver name.
If your docker daemon is configured to run with user namespace, you will need to run the container with user namespace disabled:
docker run --rm --userns host -v /var/run/docker.sock:/var/run/docker.sock -v /etc:/etc spotify/docker-gc