Merge branch '56752-replace-sidekiq-metrics' into 'master'

Adds direct monitoring for sidekiq metrics

Closes #56752

See merge request gitlab-org/gitlab-ce!30916

config/initializers/sidekiq.rb

@@ -33,6 +33,7 @@ Sidekiq.configure_server do |config|
   config.redis = queues_config_hash
 
   config.server_middleware do |chain|
+    chain.add Gitlab::SidekiqMiddleware::Metrics if Settings.monitoring.sidekiq_exporter
     chain.add Gitlab::SidekiqMiddleware::ArgumentsLogger if ENV['SIDEKIQ_LOG_ARGUMENTS'] && !enable_json_logs
     chain.add Gitlab::SidekiqMiddleware::MemoryKiller if ENV['SIDEKIQ_MEMORY_KILLER_MAX_RSS']
     chain.add Gitlab::SidekiqMiddleware::RequestStoreMiddleware unless ENV['SIDEKIQ_REQUEST_STORE'] == '0'

lib/gitlab/sidekiq_middleware/metrics.rb

+# frozen_string_literal: true
+
+module Gitlab
+  module SidekiqMiddleware
+    class Metrics
+      def initialize
+        @metrics = init_metrics
+      end
+
+      def call(_worker, job, queue)
+        labels = create_labels(queue)
+        @metrics[:sidekiq_running_jobs].increment(labels, 1)
+
+        if job['retry_count'].present?
+          @metrics[:sidekiq_jobs_retried_total].increment(labels, 1)
+        end
+
+        realtime = Benchmark.realtime do
+          yield
+        end
+
+        @metrics[:sidekiq_jobs_completion_seconds].observe(labels, realtime)
+      rescue Exception # rubocop: disable Lint/RescueException
+        @metrics[:sidekiq_jobs_failed_total].increment(labels, 1)
+        raise
+      ensure
+        @metrics[:sidekiq_running_jobs].increment(labels, -1)
+      end
+
+      private
+
+      def init_metrics
+        {
+          sidekiq_jobs_completion_seconds: ::Gitlab::Metrics.histogram(:sidekiq_jobs_completion_seconds, 'Seconds to complete sidekiq job'),
+          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'),
+          sidekiq_running_jobs:            ::Gitlab::Metrics.gauge(:sidekiq_running_jobs, 'Number of Sidekiq jobs running', {}, :livesum)
+        }
+      end
+
+      def create_labels(queue)
+        {
+          queue: queue
+        }
+      end
+    end
+  end
+end

spec/lib/gitlab/sidekiq_middleware/metrics_spec.rb

+# frozen_string_literal: true
+
+require 'spec_helper'
+
+describe Gitlab::SidekiqMiddleware::Metrics do
+  describe '#call' do
+    let(:middleware) { described_class.new }
+    let(:worker) { double(:worker) }
+
+    let(:completion_seconds_metric) { double('completion 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).and_return(completion_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)
+
+      allow(running_jobs_metric).to receive(:increment)
+    end
+
+    it 'yields block' do
+      allow(completion_seconds_metric).to receive(:observe)
+
+      expect { |b| middleware.call(worker, {}, :test, &b) }.to yield_control.once
+    end
+
+    it 'sets metrics' do
+      labels = { queue: :test }
+
+      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))
+
+      middleware.call(worker, {}, :test) { nil }
+    end
+
+    context 'when job is retried' do
+      it 'sets sidekiq_jobs_retried_total metric' do
+        allow(completion_seconds_metric).to receive(:observe)
+
+        expect(retried_total_metric).to receive(:increment)
+
+        middleware.call(worker, { 'retry_count' => 1 }, :test) { nil }
+      end
+    end
+
+    context 'when error is raised' do
+      it 'sets sidekiq_jobs_failed_total and reraises' do
+        expect(failed_total_metric).to receive(:increment)
+        expect { middleware.call(worker, {}, :test) { raise } }.to raise_error
+      end
+    end
+  end
+end