Merge branch 'avoid-timeout-in-sidekiq-queue-job-deletion' into 'master'

Avoid using Timeout when deleting Sidekiq jobs

See merge request gitlab-org/gitlab!26392

lib/gitlab/sidekiq_queue.rb

@@ -14,7 +14,8 @@ module Gitlab
     end
 
     def drop_jobs!(search_metadata, timeout:)
-      completed = false
+      start_time = Gitlab::Metrics::System.monotonic_time
+      completed = true
       deleted_jobs = 0
 
       job_search_metadata =
@@ -27,20 +28,18 @@ module Gitlab
       raise NoMetadataError if job_search_metadata.empty?
       raise InvalidQueueError unless queue
 
-      begin
-        Timeout.timeout(timeout) do
       queue.each do |job|
+        if timeout_exceeded?(start_time, timeout)
+          completed = false
+          break
+        end
+
         next unless job_matches?(job, job_search_metadata)
 
         job.delete
         deleted_jobs += 1
       end
 
-          completed = true
-        end
-      rescue Timeout::Error
-      end
-
       {
         completed: completed,
         deleted_jobs: deleted_jobs,
@@ -48,6 +47,8 @@ module Gitlab
       }
     end
 
+    private
+
     def queue
       strong_memoize(:queue) do
         # Sidekiq::Queue.new always returns a queue, even if it doesn't
@@ -59,5 +60,9 @@ module Gitlab
     def job_matches?(job, job_search_metadata)
       job_search_metadata.all? { |key, value| job[key] == value }
     end
+
+    def timeout_exceeded?(start_time, timeout)
+      (Gitlab::Metrics::System.monotonic_time - start_time) > timeout
+    end
   end
 end

spec/lib/gitlab/sidekiq_queue_spec.rb

@@ -31,14 +31,7 @@ describe Gitlab::SidekiqQueue do
 
       context 'when the queue is not processed in time' do
         before do
-          calls = 0
-
-          allow(sidekiq_queue).to receive(:job_matches?).and_wrap_original do |m, *args|
-            raise Timeout::Error if calls > 0
-
-            calls += 1
-            m.call(*args)
-          end
+          allow(Gitlab::Metrics::System).to receive(:monotonic_time).and_return(1, 2, 12)
         end
 
         it 'returns a non-completion flag, the number of jobs deleted, and the remaining queue size' do