Remove expired artifacts periodically

Rename

Introduce Destroy expired job artifacts service

Revert a bit

Add changelog

Use expired

Improve

Fix spec

Fix spec

Use bang for destroy

Introduce iteration limit

Update comment

Simplify more

Refacor

Remove unnecessary thing

Fix comments

Fix coding offence

Make loop helper exception free

3 files changed, 53 insertions, 1 deletions

diff --git a/app/models/ci/job_artifact.rb b/app/models/ci/job_artifact.rb
index 11c88200c37..789bb293811 100644
--- a/app/models/ci/job_artifact.rb
+++ b/app/models/ci/job_artifact.rb
@@ -73,6 +73,8 @@ module Ci
       where(file_type: types)
     end
 
+    scope :expired, -> (limit) { where('expire_at < ?', Time.now).limit(limit) }
+
     delegate :filename, :exists?, :open, to: :file
 
     enum file_type: {
diff --git a/app/services/ci/destroy_expired_job_artifacts_service.rb b/app/services/ci/destroy_expired_job_artifacts_service.rb
new file mode 100644
index 00000000000..7d2f5d33fed
--- /dev/null
+++ b/app/services/ci/destroy_expired_job_artifacts_service.rb
@@ -0,0 +1,38 @@
+# frozen_string_literal: true
+
+module Ci
+  class DestroyExpiredJobArtifactsService
+    include ::Gitlab::ExclusiveLeaseHelpers
+    include ::Gitlab::LoopHelpers
+
+    BATCH_SIZE = 100
+    LOOP_TIMEOUT = 45.minutes
+    LOOP_LIMIT = 1000
+    EXCLUSIVE_LOCK_KEY = 'expired_job_artifacts:destroy:lock'
+    LOCK_TIMEOUT = 50.minutes
+
+    ##
+    # Destroy expired job artifacts on GitLab instance
+    #
+    # This destroy process cannot run for more than 45 minutes. This is for
+    # preventing multiple `ExpireBuildArtifactsWorker` CRON jobs run concurrently,
+    # which is scheduled at every hour.
+    def execute
+      in_lock(EXCLUSIVE_LOCK_KEY, ttl: LOCK_TIMEOUT, retries: 1) do
+        loop_until(timeout: LOOP_TIMEOUT, limit: LOOP_LIMIT) do
+          destroy_batch
+        end
+      end
+    end
+
+    private
+
+    def destroy_batch
+      artifacts = Ci::JobArtifact.expired(BATCH_SIZE).to_a
+
+      return false if artifacts.empty?
+
+      artifacts.each(&:destroy!)
+    end
+  end
+end
diff --git a/app/workers/expire_build_artifacts_worker.rb b/app/workers/expire_build_artifacts_worker.rb
index dce812d1ae2..251e95c68d5 100644
--- a/app/workers/expire_build_artifacts_worker.rb
+++ b/app/workers/expire_build_artifacts_worker.rb
@@ -4,8 +4,20 @@ class ExpireBuildArtifactsWorker
   include ApplicationWorker
   include CronjobQueue
 
-  # rubocop: disable CodeReuse/ActiveRecord
   def perform
+    if Feature.enabled?(:ci_new_expire_job_artifacts_service, default_enabled: true)
+      perform_efficient_artifacts_removal
+    else
+      perform_legacy_artifacts_removal
+    end
+  end
+
+  def perform_efficient_artifacts_removal
+    Ci::DestroyExpiredJobArtifactsService.new.execute
+  end
+
+  # rubocop: disable CodeReuse/ActiveRecord
+  def perform_legacy_artifacts_removal
     Rails.logger.info 'Scheduling removal of build artifacts'
 
     build_ids = Ci::Build.with_expired_artifacts.pluck(:id)