A script to remove orphaned docker images in a docker registry. Ported to Ruby for easy of understanding.

.rubocop.yml

Style/MultilineOperationIndentation:
  EnforcedStyle: indented

Style/FileName:
  Enabled: false

Metrics/AbcSize:
  Enabled: false

Metrics/LineLength:
  Max: 120

README.md

docker-registry-garbage-collect

The new home page is docwhat/docker-registry-garbage-collect on github.

Docker Registery version 1 has one huge glaring problem: It doesn't clean up "dangling" or "orphaned" images.

This happens when you push up "foobar:latest" and it clobbers a previous "foobar:latest". The images that made up the previous versions are now orphaned and left floating in limbo.

Notes

For maximum paranoia, the script will not remove images from the last hour. This is in case you're uploading something that hasn't been tagged yet.

License

Public domain. This stuff is trivial and it is needed.

Credits

• The original shell script was done by shepmaster.
• bjaglin created a second version shell script.
• The final shell script, which I ported here, was by kwk.

remove-orphan-images.rb

#!/usr/bin/env ruby

require 'optparse'
require 'pathname'
require 'json'
require 'fileutils'

# Cleans up all images not referenced by a tag.
class Application
  attr_reader :base_dir

  def initialize(args)
    @base_dir = Pathname.new('/data/registry')
    @args = args
  end

  def log_stream
    @log_stream ||= File.open('/tmp/docker-remove-orphan-images.log', 'a')
      .tap { |l| l.puts "*** STARTING AT #{Time.now} ***" }
  end

  def info(str)
    log_stream.puts "INFO: #{str}"
    puts "INFO: #{str}"
  end

  def repository_dir
    base_dir + 'repositories'
  end

  def image_dir
    base_dir + 'images'
  end

  def libraries
    @libraries ||= repository_dir
      .children
      .select(&:directory?)
  end

  def repositories
    @repositories ||= libraries
      .map(&:children)
      .flatten
      .select(&:directory?)
  end

  def tags
    @tags ||= repositories
      .map(&:children)
      .flatten
      .select { |p| p.basename.to_s =~ /^tag_/ }
  end

  def all_image_hashes
    @all_image_hashes ||= image_dir
      .children
      .select(&:directory?)
      .select { |p| (p + '_checksum').exist? }
      .reject { |p| (p + '_checksum').mtime >= timestamp }
      .map(&:basename)
      .map(&:to_s)
  end

  def used_image_hashes
    @used_image_hashes ||= tags
      .map(&:read)
      .map(&:chomp)
      .map { |h| image_dir + h + 'ancestry' }
      .select(&:file?)
      .map { |p| JSON.load p }
      .flatten
      .sort
      .uniq
  end

  def unused_image_hashes
    @unused_image_hashes ||= all_image_hashes - used_image_hashes
  end

  def timestamp
    @timestamp ||= Time.new - (60 * 60) # in seconds
  end

  def remove_index_references!
    repositories.each do |repo|
      filename = repo + '_index_images'
      original_index = JSON.load filename.read
      modified_index = original_index.reject { |x| unused_image_hashes.include? x['id'] }
      next if original_index.size == modified_index.size

      info "Updating index for #{repo}"
      filename.open('w') { |f| f.write modified_index.to_json }
    end
  end

  def remove_unused_images!
    unused_image_hashes.each do |hash|
      filename = image_dir + hash
      info "Removing #{hash}"
      FileUtils.rm_rf filename
    end
  end

  def run
    remove_index_references!
    remove_unused_images!
  ensure
    log_stream.close unless log_stream.closed?
  end
end

Application.new(ARGV).run if $PROGRAM_NAME == __FILE__