Yet another Docker GC

Yet another Docker GC but unlike others :

• Written in Go and uses the go-dockerclient to talk straight to API (rather than shelling out)
• Actually have tests
• Supports manual cleanups in addition to continuous runs
• Supports reporting errors to Bugsnag
• Supports sending metrics to dogstatsd

Usage

Compiling

You can compile the binary by doing script/setup and script/compile. Golang 1.5+ and Git 1.7+ is needed

Running

  docker-gc (-command=containers|images|all|emergency) (-keep_last_images=DURATION) (-keep_last_containers=DURATION)
  -command=all cleans all images and containes respecting keep_last values
  -command=emergency same as all, but with 0second keep_last values
  OR
  docker-gc (-command=continuous) (-interval=INTERVAL_IN_SECONDS) (-keep_last_images=DURATION) (-keep_last_containers=DURATION) for continuous cleanup in TTL mode
  OR
  docker-gc (-command=diskspace) (-interval=INTERVAL_IN_SECONDS) (-high_disk_space_threshold=PERCENTAGE) (-low_disk_space_threshold=PERCENTAGE) for disk space based continuous mode

  You can also specify -bugsnag-key="key" to use bugsnag integration
  and -statsd_address=127.0.0.1:815 and statsd_namespace=docker.gc.wtf. for statsd integration

docker-gc has two main modes; continuous cleanup and one-time cleanup.

One-time cleanup

One-time cleanup can be in four ways (as -command=COMMAND)

• emergency : clean all containers and images
• all : clean all containers and images but respect keep_last values
• images/containers : clean only images or only containers respecting keep_last values

eg. docker-gc -command=all -keep_last_images=5m -keep_last_containers=1m would do a one time cleanup of images older than 5minutes and containers older than 1minutes

Default values are:

• command = continuous
• keep_last_images = 10 hours
• keep_last_containers = 1minutes

Continuous mode

docker-gc has two continuous modes; TTL based and free disk space based. This means the daemon keeps running and does swipes per interval settings.

Default value for interval is 60 seconds.

TTL based

eg docker-gc -command=continuous -interval=5m

Same TTL settings for containers and images apply than for One-time cleanup

Free disk space based

eg docker-gc -command=diskspace -interval=5m -high_disk_space_threshold=85 -low_disk_space_threshold=50

Monitors the disk volume used by Docker and if used disk space hits the high_disk_space_threshold threshold starts cleaning up containers and images in batches of 10 until low_disk_space_threshold is reached.

NOTICE: for containers we cleanup based on the keep_last_containers because in majority of usecases it makes more senses than looping in batches. For images we do this in patches of 10 starting from the oldest after that.

Usage

Development can be done on both OSX and Linux. Tests can be run without Docker, but anykind of manual testing requires your user to have rights to unix:///var/run/docker.sock (eg. be in docker group)

Setup your environment by doing script/setup. Tests can be run with script/test and running the current version with script/run (compile+run+test mode env)