Add rake task to clean orphan artifact files

This adds the rake task rake
gitlab:cleanup:orphan_job_artifact_files. This rake task cleans all
orphan job artifact files it can find on disk.

 It performs a search on the complete folder of all artifacts on
disk. Then it filters out all the job artifact ID for which it could
not find a record with matching ID in the database. For these, the
file is deleted from disk.

changelogs/unreleased/tc-rake-orphan-artifacts.yml

+---
+title: Add rake task to clean orphan artifact files
+merge_request: 29681
+author:
+type: added

doc/administration/geo/replication/troubleshooting.md

@@ -504,6 +504,15 @@ To resolve this, run the following command:
 sudo gitlab-rake geo:db:refresh_foreign_tables
 ```
+## Expired artifacts
+
+If you notice for some reason there are more artifacts on the Geo
+secondary node than on the Geo primary node, you can use the rake task
+to [cleanup orphan artifact files](../../../raketasks/cleanup.md#remove-orphan-artifact-files).
+
+On a Geo **secondary** node, this command will also clean up all Geo
+registry record related to the orphan files on disk.
+
 ## Fixing common errors
 This section documents common errors reported in the Admin UI and how to fix them.

doc/raketasks/cleanup.md

@@ -92,3 +92,48 @@ I, [2018-08-02T10:26:47.598424 #45087]  INFO -- : Looking for orphaned remote up
 I, [2018-08-02T10:26:47.753131 #45087]  INFO -- : Moved to lost and found: @hashed/6b/DSC_6152.JPG -> lost_and_found/@hashed/6b/DSC_6152.JPG
 I, [2018-08-02T10:26:47.764356 #45087]  INFO -- : Moved to lost and found: @hashed/79/02/7902699be42c8a8e46fbbb4501726517e86b22c56a189f7625a6da49081b2451/711491b29d3eb08837798c4909e2aa4d/DSC00314.jpg -> lost_and_found/@hashed/79/02/7902699be42c8a8e46fbbb4501726517e86b22c56a189f7625a6da49081b2451/711491b29d3eb08837798c4909e2aa4d/DSC00314.jpg
 ```
+
+## Remove orphan artifact files
+
+When you notice there are more job artifacts files on disk than there
+should be, you can run:
+
+```shell
+gitlab-rake gitlab:cleanup:orphan_job_artifact_files
+```
+
+This command:
+
+- Scans through the entire artifacts folder.
+- Checks which files still have a record in the database.
+- If no database record is found, the file is deleted from disk.
+
+By default, this task does not delete anything but shows what it can
+delete. Run the command with `DRY_RUN=false` if you actually want to
+delete the files:
+
+```shell
+gitlab-rake gitlab:cleanup:orphan_job_artifact_files DRY_RUN=false
+```
+
+You can also limit the number of files to delete with `LIMIT`:
+
+```shell
+gitlab-rake gitlab:cleanup:orphan_job_artifact_files LIMIT=100`
+```
+
+This will only delete up to 100 files from disk. You can use this to
+delete a small set for testing purposes.
+
+If you provide `DEBUG=1`, you'll see the full path of every file that
+is detected as being an orphan.
+
+If `ionice` is installed, the tasks uses it to ensure the command is
+not causing too much load on the disk. You can configure the niceness
+level with `NICENESS`. Below are the valid levels, but consult
+`man 1 ionice` to be sure.
+
+- `0` or `None`
+- `1` or `Realtime`
+- `2` or `Best-effort` (default)
+- `3` or `Idle`

lib/gitlab/cleanup/orphan_job_artifact_files.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module Cleanup
+    class OrphanJobArtifactFiles
+      include Gitlab::Utils::StrongMemoize
+
+      ABSOLUTE_ARTIFACT_DIR = ::JobArtifactUploader.root.freeze
+      LOST_AND_FOUND = File.join(ABSOLUTE_ARTIFACT_DIR, '-', 'lost+found').freeze
+      BATCH_SIZE = 500
+      DEFAULT_NICENESS = 'Best-effort'
+
+      attr_accessor :batch, :total_found, :total_cleaned
+      attr_reader :limit, :dry_run, :niceness, :logger
+
+      def initialize(limit: nil, dry_run: true, niceness: nil, logger: nil)
+        @limit = limit
+        @dry_run = dry_run
+        @niceness = niceness || DEFAULT_NICENESS
+        @logger = logger || Rails.logger
+        @total_found = @total_cleaned = 0
+
+        new_batch!
+      end
+
+      def run!
+        log_info('Looking for orphan job artifacts to clean up')
+
+        find_artifacts do |artifact_file|
+          batch << artifact_file
+
+          clean_batch! if batch.full?
+          break if limit_reached?
+        end
+
+        clean_batch!
+
+        log_info("Processed #{total_found} job artifacts to find and clean #{total_cleaned} orphans.")
+      end
+
+      private
+
+      def new_batch!
+        self.batch = ::Gitlab::Cleanup::OrphanJobArtifactFilesBatch
+          .new(batch_size: batch_size, logger: logger, dry_run: dry_run)
+      end
+
+      def clean_batch!
+        batch.clean!
+
+        update_stats!(batch)
+
+        new_batch!
+      end
+
+      def update_stats!(batch)
+        self.total_found += batch.artifact_files.count
+        self.total_cleaned += batch.lost_and_found.count
+      end
+
+      def limit_reached?
+        return false unless limit
+
+        total_cleaned >= limit
+      end
+
+      def batch_size
+        return BATCH_SIZE unless limit
+        return if limit_reached?
+
+        todo = limit - total_cleaned
+        [BATCH_SIZE, todo].min
+      end
+
+      def find_artifacts
+        Open3.popen3(*find_command) do |stdin, stdout, stderr, status_thread|
+          stdout.each_line do |line|
+            yield line
+          end
+
+          log_error(stderr.read.color(:red)) unless status_thread.value.success?
+        end
+      end
+
+      def find_command
+        strong_memoize(:find_command) do
+          cmd = %W[find -L #{absolute_artifact_dir}]
+
+          # Search for Job Artifact IDs, they are found 6 directory
+          # levels deep. For example:
+          # shared/artifacts/2c/62/2c...a3/2019_02_27/836/628/job.log
+          #                  1  2  3       4          5   6
+          #                  |  |  |       ^- date    |   ^- Job Artifact ID
+          #                  |  |  |                  ^- Job ID
+          #                  ^--+--+- components of hashed storage project path
+          cmd += %w[-mindepth 6 -maxdepth 6]
+
+          # Artifact directories are named on their ID
+          cmd += %w[-type d]
+
+          if ionice
+            raise ArgumentError, 'Invalid niceness' unless niceness.match?(/^\w[\w\-]*$/)
+
+            cmd.unshift(*%W[#{ionice} --class #{niceness}])
+          end
+
+          log_info("find command: '#{cmd.join(' ')}'")
+
+          cmd
+        end
+      end
+
+      def absolute_artifact_dir
+        File.absolute_path(ABSOLUTE_ARTIFACT_DIR)
+      end
+
+      def ionice
+        strong_memoize(:ionice) do
+          Gitlab::Utils.which('ionice')
+        end
+      end
+
+      def log_info(msg, params = {})
+        logger.info("#{'[DRY RUN]' if dry_run} #{msg}")
+      end
+
+      def log_error(msg, params = {})
+        logger.error(msg)
+      end
+    end
+  end
+end

lib/gitlab/cleanup/orphan_job_artifact_files_batch.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module Cleanup
+    class OrphanJobArtifactFilesBatch
+      BatchFull = Class.new(StandardError)
+
+      class ArtifactFile
+        attr_accessor :path
+
+        def initialize(path)
+          @path = path
+        end
+
+        def artifact_id
+          path.split('/').last.to_i
+        end
+      end
+
+      include Gitlab::Utils::StrongMemoize
+
+      attr_reader :batch_size, :dry_run
+      attr_accessor :artifact_files
+
+      def initialize(batch_size:, dry_run: true, logger: Rails.logger)
+        @batch_size = batch_size
+        @dry_run = dry_run
+        @logger = logger
+        @artifact_files = []
+      end
+
+      def clean!
+        return if artifact_files.empty?
+
+        lost_and_found.each do |artifact|
+          clean_one!(artifact)
+        end
+      end
+
+      def full?
+        artifact_files.count >= batch_size
+      end
+
+      def <<(artifact_path)
+        raise BatchFull, "Batch full! Already contains #{artifact_files.count} artifacts" if full?
+
+        artifact_files << ArtifactFile.new(artifact_path)
+      end
+
+      def lost_and_found
+        strong_memoize(:lost_and_found) do
+          artifact_file_ids = artifact_files.map(&:artifact_id)
+          existing_artifact_ids = ::Ci::JobArtifact.id_in(artifact_file_ids).pluck_primary_key
+
+          artifact_files.reject { |artifact| existing_artifact_ids.include?(artifact.artifact_id) }
+        end
+      end
+
+      private
+
+      def clean_one!(artifact_file)
+        log_debug("Found orphan job artifact file @ #{artifact_file.path}")
+
+        remove_file!(artifact_file) unless dry_run
+      end
+
+      def remove_file!(artifact_file)
+        FileUtils.rm_rf(artifact_file.path)
+      end
+
+      def log_info(msg, params = {})
+        @logger.info("#{'[DRY RUN]' if dry_run} #{msg}")
+      end
+
+      def log_debug(msg, params = {})
+        @logger.debug(msg)
+      end
+    end
+  end
+end

lib/tasks/gitlab/cleanup.rake

@@ -115,6 +115,18 @@ namespace :gitlab do
       end
     end
+    desc 'GitLab | Cleanup | Clean orphan job artifact files'
+    task orphan_job_artifact_files: :gitlab_environment do
+      warn_user_is_not_gitlab
+
+      cleaner = Gitlab::Cleanup::OrphanJobArtifactFiles.new(limit: limit, dry_run: dry_run?, niceness: niceness, logger: logger)
+      cleaner.run!
+
+      if dry_run?
+        logger.info "To clean up these files run this command with DRY_RUN=false".color(:yellow)
+      end
+    end
+
     def remove?
       ENV['REMOVE'] == 'true'
     end
@@ -123,12 +135,25 @@ namespace :gitlab do
       ENV['DRY_RUN'] != 'false'
     end
+    def debug?
+      ENV['DEBUG'].present?
+    end
+
+    def limit
+      ENV['LIMIT']&.to_i
+    end
+
+    def niceness
+      ENV['NICENESS'].presence
+    end
+
     def logger
       return @logger if defined?(@logger)
       @logger = if Rails.env.development? || Rails.env.production?
                   Logger.new(STDOUT).tap do |stdout_logger|
                     stdout_logger.extend(ActiveSupport::Logger.broadcast(Rails.logger))
+                    stdout_logger.level = debug? ? Logger::DEBUG : Logger::INFO
                   end
                 else
                   Rails.logger

spec/lib/gitlab/cleanup/orphan_job_artifact_files_batch_spec.rb

+# frozen_string_literal: true
+
+require 'spec_helper'
+
+describe Gitlab::Cleanup::OrphanJobArtifactFilesBatch do
+  let(:batch_size) { 10 }
+  let(:dry_run) { true }
+
+  subject(:batch) { described_class.new(batch_size: batch_size, dry_run: dry_run) }
+
+  context 'no dry run' do
+    let(:dry_run) { false }
+
+    it 'deletes only orphan job artifacts from disk' do
+      job_artifact = create(:ci_job_artifact, :archive)
+      orphan_artifact = create(:ci_job_artifact, :archive)
+      batch << artifact_path(job_artifact)
+      batch << artifact_path(orphan_artifact)
+      orphan_artifact.delete
+
+      batch.clean!
+
+      expect(batch.artifact_files.count).to eq(2)
+      expect(batch.lost_and_found.count).to eq(1)
+      expect(batch.lost_and_found.first.artifact_id).to eq(orphan_artifact.id)
+    end
+
+    it 'does not mix up job ID and artifact ID' do
+      # take maximum ID of both tables to avoid any collision
+      max_id = [Ci::Build.maximum(:id), Ci::JobArtifact.maximum(:id)].compact.max.to_i
+      job_a = create(:ci_build, id: max_id + 1)
+      job_b = create(:ci_build, id: max_id + 2)
+      # reuse the build IDs for the job artifact IDs, but swap them
+      job_artifact_b = create(:ci_job_artifact, :archive, job: job_b, id: max_id + 1)
+      job_artifact_a = create(:ci_job_artifact, :archive, job: job_a, id: max_id + 2)
+
+      batch << artifact_path(job_artifact_a)
+      batch << artifact_path(job_artifact_b)
+
+      job_artifact_b.delete
+
+      batch.clean!
+
+      expect(File.exist?(job_artifact_a.file.path)).to be_truthy
+      expect(File.exist?(job_artifact_b.file.path)).to be_falsey
+    end
+  end
+
+  context 'with dry run' do
+    it 'does not remove files' do
+      job_artifact = create(:ci_job_artifact, :archive)
+      batch << job_artifact.file.path
+      job_artifact.delete
+
+      expect(batch).not_to receive(:remove_file!)
+
+      batch.clean!
+
+      expect(File.exist?(job_artifact.file.path)).to be_truthy
+    end
+  end
+
+  def artifact_path(job_artifact)
+    Pathname.new(job_artifact.file.path).parent.to_s
+  end
+end

spec/lib/gitlab/cleanup/orphan_job_artifact_files_spec.rb

+# frozen_string_literal: true
+
+require 'spec_helper'
+
+describe Gitlab::Cleanup::OrphanJobArtifactFiles do
+  let(:null_logger) { Logger.new('/dev/null') }
+  subject(:cleanup) { described_class.new(logger: null_logger) }
+
+  before do
+    allow(null_logger).to receive(:info)
+  end
+
+  it 'passes on dry_run' do
+    expect(Gitlab::Cleanup::OrphanJobArtifactFilesBatch)
+      .to receive(:new)
+      .with(dry_run: false, batch_size: anything, logger: anything)
+      .at_least(:once)
+      .and_call_original
+
+    described_class.new(dry_run: false).run!
+  end
+
+  it 'errors when invalid niceness is given' do
+    cleanup = described_class.new(logger: null_logger, niceness: 'FooBar')
+
+    expect(null_logger).to receive(:error).with(/FooBar/)
+
+    cleanup.run!
+  end
+
+  it 'finds artifacts on disk' do
+    artifact = create(:ci_job_artifact, :archive)
+
+    expect(cleanup).to receive(:find_artifacts).and_yield(artifact.file.path)
+    cleanup.run!
+  end
+
+  it 'stops when limit is reached' do
+    cleanup = described_class.new(limit: 1)
+
+    mock_artifacts_found(cleanup, 'tmp/foo/bar/1', 'tmp/foo/bar/2')
+
+    cleanup.run!
+
+    expect(cleanup.total_found).to eq(1)
+  end
+
+  it 'cleans even if batch is not full' do
+    mock_artifacts_found(cleanup, 'tmp/foo/bar/1')
+
+    expect(cleanup).to receive(:clean_batch!).and_call_original
+    cleanup.run!
+  end
+
+  it 'cleans in batches' do
+    stub_const("#{described_class.name}::BATCH_SIZE", 2)
+    mock_artifacts_found(cleanup, 'tmp/foo/bar/1', 'tmp/foo/bar/2', 'tmp/foo/bar/3')
+
+    expect(cleanup).to receive(:clean_batch!).twice.and_call_original
+    cleanup.run!
+  end
+
+  def mock_artifacts_found(cleanup, *files)
+    mock = allow(cleanup).to receive(:find_artifacts)
+
+    files.each { |file| mock.and_yield(file) }
+  end
+end

spec/tasks/gitlab/cleanup_rake_spec.rb

@@ -156,4 +156,33 @@ describe 'gitlab:cleanup rake tasks' do
       end
     end
   end
+
+  describe 'gitlab:cleanup:orphan_job_artifact_files' do
+    subject(:rake_task) { run_rake_task('gitlab:cleanup:orphan_job_artifact_files') }
+
+    it 'runs the task without errors' do
+      expect(Gitlab::Cleanup::OrphanJobArtifactFiles)
+        .to receive(:new).and_call_original
+
+      expect { rake_task }.not_to raise_error
+    end
+
+    context 'with DRY_RUN set to false' do
+      before do
+        stub_env('DRY_RUN', 'false')
+      end
+
+      it 'passes dry_run correctly' do
+        expect(Gitlab::Cleanup::OrphanJobArtifactFiles)
+          .to receive(:new)
+          .with(limit: anything,
+                dry_run: false,
+                niceness: anything,
+                logger: anything)
+          .and_call_original
+
+        rake_task
+      end
+    end
+  end
 end