Allow capture of error class into tag when job fails (#34)

lib/yabeda/sidekiq.rb

@@ -29,9 +29,11 @@ def self.config
       counter :jobs_rerouted_total, tags: %i[from_queue to_queue worker], comment: "A counter of the total number of rerouted jobs sidekiq enqueued."
 
       if config.declare_process_metrics # defaults to +::Sidekiq.server?+
+        failed_total_tags = config.label_for_error_class_on_sidekiq_jobs_failed ? %i[queue worker error] : %i[queue worker]
+
         counter   :jobs_executed_total,  tags: %i[queue worker], comment: "A counter of the total number of jobs sidekiq executed."
         counter   :jobs_success_total,   tags: %i[queue worker], comment: "A counter of the total number of jobs successfully processed by sidekiq."
-        counter   :jobs_failed_total,    tags: %i[queue worker], comment: "A counter of the total number of jobs failed in sidekiq."
+        counter   :jobs_failed_total,    tags: failed_total_tags, comment: "A counter of the total number of jobs failed in sidekiq."
 
         gauge     :running_job_runtime,  tags: %i[queue worker], aggregation: :max, unit: :seconds,
                                          comment: "How long currently running jobs are running (useful for detection of hung jobs)"

lib/yabeda/sidekiq/config.rb

@@ -20,6 +20,9 @@ class Config < ::Anyway::Config
       # Retries are tracked by default as a single metric. If you want to track them separately for each queue, set this to +true+
       # Disabled by default because it is quite slow if the retry set is large
       attr_config retries_segmented_by_queue: false
+
+      # If set to true, an `:error` label will be added with name of the error class to all failed jobs
+      attr_config label_for_error_class_on_sidekiq_jobs_failed: false
     end
   end
 end

lib/yabeda/sidekiq/server_middleware.rb

@@ -20,8 +20,10 @@ def call(worker, job, queue)
             yield
           end
           Yabeda.sidekiq_jobs_success_total.increment(labels)
-        rescue Exception # rubocop: disable Lint/RescueException
-          Yabeda.sidekiq_jobs_failed_total.increment(labels)
+        rescue Exception => e # rubocop: disable Lint/RescueException
+          jobs_failed_labels = labels.dup
+          jobs_failed_labels[:error] = e.class.name if Yabeda::Sidekiq.config.label_for_error_class_on_sidekiq_jobs_failed
+          Yabeda.sidekiq_jobs_failed_total.increment(jobs_failed_labels)
           raise
         ensure
           Yabeda.sidekiq_job_runtime.measure(labels, elapsed(start))

spec/support/jobs.rb

@@ -33,8 +33,10 @@ def yabeda_tags
 class FailingPlainJob
   include Sidekiq::Worker
 
+  SpecialError = Class.new(StandardError)
+
   def perform(*_args)
-    raise "Badaboom"
+    raise SpecialError, "Badaboom"
   end
 end
 
@@ -47,9 +49,11 @@ def perform(*_args)
 end
 
 class FailingActiveJob < ActiveJob::Base
+  SpecialError = Class.new(StandardError)
+
   self.queue_adapter = :Sidekiq
   def perform(*_args)
-    raise "Boom"
+    raise SpecialError, "Boom"
   end
 end
 

spec/yabeda/sidekiq_spec.rb

@@ -23,6 +23,51 @@
         )
     end
 
+    context "when label_for_error_class_on_sidekiq_jobs_failed is set to true" do
+      around do |example|
+        old_value = described_class.config.label_for_error_class_on_sidekiq_jobs_failed
+        described_class.config.label_for_error_class_on_sidekiq_jobs_failed = true
+
+        example.run
+
+        described_class.config.label_for_error_class_on_sidekiq_jobs_failed = old_value
+      end
+
+      it "counts failed total and executed total with correct labels", sidekiq: :inline do
+        expect do
+          SamplePlainJob.perform_async
+          SamplePlainJob.perform_async
+          begin
+            FailingPlainJob.perform_async
+          rescue StandardError
+            nil
+          end
+        end.to \
+          increment_yabeda_counter(Yabeda.sidekiq.jobs_failed_total).with(
+            { queue: "default", worker: "FailingPlainJob", error: "FailingPlainJob::SpecialError" } => 1,
+          ).and \
+            increment_yabeda_counter(Yabeda.sidekiq.jobs_executed_total).with(
+              { queue: "default", worker: "SamplePlainJob" } => 2,
+              { queue: "default", worker: "FailingPlainJob" } => 1,
+            )
+      end
+
+      it "does not add jobs_failed_total error label to labels used for jobs_executed_total", sidekiq: :inline do
+        expect do
+          SamplePlainJob.perform_async
+          SamplePlainJob.perform_async
+          begin
+            FailingPlainJob.perform_async
+          rescue StandardError
+            nil
+          end
+        end.not_to \
+          increment_yabeda_counter(Yabeda.sidekiq.jobs_executed_total).with(
+            { queue: "default", worker: "FailingPlainJob", error: "FailingPlainJob::SpecialError" } => 1,
+          )
+      end
+    end
+
     describe "re-routing jobs by middleware" do
       around do |example|
         add_reroute_jobs_middleware
@@ -100,6 +145,32 @@
       end
     end
 
+    context "when label_for_error_class_on_sidekiq_jobs_failed is set to true" do
+      around do |example|
+        old_value = described_class.config.label_for_error_class_on_sidekiq_jobs_failed
+        described_class.config.label_for_error_class_on_sidekiq_jobs_failed = true
+
+        example.run
+
+        described_class.config.label_for_error_class_on_sidekiq_jobs_failed = old_value
+      end
+
+      it "counts enqueues and uses the default label for the error class", sidekiq: :inline do
+        expect do
+          SampleActiveJob.perform_later
+          SampleActiveJob.perform_later
+          begin
+            FailingActiveJob.perform_later
+          rescue StandardError
+            nil
+          end
+        end.to \
+          increment_yabeda_counter(Yabeda.sidekiq.jobs_failed_total).with(
+            { queue: "default", worker: "FailingActiveJob", error: "FailingActiveJob::SpecialError" } => 1,
+          )
+      end
+    end
+
     it "measures runtime", sidekiq: :inline do
       expect do
         SampleActiveJob.perform_later