Merge branch 'sidekiq-job-cputime-prometheus-metrics' into 'master'

Add Sidekiq job CPU time prometheus metrics

See merge request gitlab-org/gitlab!17682

lib/gitlab/sidekiq_middleware/metrics.rb

@@ -19,10 +19,16 @@ module Gitlab
           @metrics[:sidekiq_jobs_retried_total].increment(labels, 1)
         end
 
+        job_thread_cputime_start = get_thread_cputime
+
         realtime = Benchmark.realtime do
           yield
         end
 
+        job_thread_cputime_end = get_thread_cputime
+        job_thread_cputime = job_thread_cputime_end - job_thread_cputime_start
+        @metrics[:sidekiq_jobs_cpu_seconds].observe(labels, job_thread_cputime)
+
         @metrics[:sidekiq_jobs_completion_seconds].observe(labels, realtime)
       rescue Exception # rubocop: disable Lint/RescueException
         @metrics[:sidekiq_jobs_failed_total].increment(labels, 1)
@@ -35,6 +41,7 @@ module Gitlab
 
       def init_metrics
         {
+          sidekiq_jobs_cpu_seconds: ::Gitlab::Metrics.histogram(:sidekiq_jobs_cpu_seconds, 'Seconds of cpu time to run sidekiq job', {}, SIDEKIQ_LATENCY_BUCKETS),
           sidekiq_jobs_completion_seconds: ::Gitlab::Metrics.histogram(:sidekiq_jobs_completion_seconds, 'Seconds to complete sidekiq job', {}, SIDEKIQ_LATENCY_BUCKETS),
           sidekiq_jobs_failed_total:       ::Gitlab::Metrics.counter(:sidekiq_jobs_failed_total, 'Sidekiq jobs failed'),
           sidekiq_jobs_retried_total:      ::Gitlab::Metrics.counter(:sidekiq_jobs_retried_total, 'Sidekiq jobs retried'),
@@ -47,6 +54,10 @@ module Gitlab
           queue: queue
         }
       end
+
+      def get_thread_cputime
+        defined?(Process::CLOCK_THREAD_CPUTIME_ID) ? Process.clock_gettime(Process::CLOCK_THREAD_CPUTIME_ID) : 0
+      end
     end
   end
 end

spec/lib/gitlab/sidekiq_middleware/metrics_spec.rb

@@ -8,12 +8,14 @@ describe Gitlab::SidekiqMiddleware::Metrics do
     let(:worker) { double(:worker) }
 
     let(:completion_seconds_metric) { double('completion seconds metric') }
+    let(:user_execution_seconds_metric) { double('user execution seconds metric') }
     let(:failed_total_metric) { double('failed total metric') }
     let(:retried_total_metric) { double('retried total metric') }
     let(:running_jobs_metric) { double('running jobs metric') }
 
     before do
       allow(Gitlab::Metrics).to receive(:histogram).with(:sidekiq_jobs_completion_seconds, anything, anything, anything).and_return(completion_seconds_metric)
+      allow(Gitlab::Metrics).to receive(:histogram).with(:sidekiq_jobs_cpu_seconds, anything, anything, anything).and_return(user_execution_seconds_metric)
       allow(Gitlab::Metrics).to receive(:counter).with(:sidekiq_jobs_failed_total, anything).and_return(failed_total_metric)
       allow(Gitlab::Metrics).to receive(:counter).with(:sidekiq_jobs_retried_total, anything).and_return(retried_total_metric)
       allow(Gitlab::Metrics).to receive(:gauge).with(:sidekiq_running_jobs, anything, {}, :livesum).and_return(running_jobs_metric)
@@ -23,13 +25,16 @@ describe Gitlab::SidekiqMiddleware::Metrics do
 
     it 'yields block' do
       allow(completion_seconds_metric).to receive(:observe)
+      allow(user_execution_seconds_metric).to receive(:observe)
 
       expect { |b| middleware.call(worker, {}, :test, &b) }.to yield_control.once
     end
 
     it 'sets metrics' do
       labels = { queue: :test }
+      allow(middleware).to receive(:get_thread_cputime).and_return(1, 3)
 
+      expect(user_execution_seconds_metric).to receive(:observe).with(labels, 2)
       expect(running_jobs_metric).to receive(:increment).with(labels, 1)
       expect(running_jobs_metric).to receive(:increment).with(labels, -1)
       expect(completion_seconds_metric).to receive(:observe).with(labels, kind_of(Numeric))
@@ -37,9 +42,17 @@ describe Gitlab::SidekiqMiddleware::Metrics do
       middleware.call(worker, {}, :test) { nil }
     end
 
+    it 'ignore user execution when measured 0' do
+      allow(completion_seconds_metric).to receive(:observe)
+      allow(middleware).to receive(:get_thread_cputime).and_return(0, 0)
+
+      expect(user_execution_seconds_metric).not_to receive(:observe)
+    end
+
     context 'when job is retried' do
       it 'sets sidekiq_jobs_retried_total metric' do
         allow(completion_seconds_metric).to receive(:observe)
+        expect(user_execution_seconds_metric).to receive(:observe)
 
         expect(retried_total_metric).to receive(:increment)