feat: metrics reporter for Jaeger collector exporter (#1271)

* tidy: move tests directory to match lib structure

* feat: add support for a metrics reporter

* fix: remove duplicated test

* chore: switch to rspec mocks

Co-authored-by: Francis Bogsanyi <[email protected]>

Author: chrisholmes

exporter/jaeger/lib/opentelemetry/exporter/jaeger/collector_exporter.rb

@@ -29,7 +29,8 @@ def initialize(endpoint: ENV.fetch('OTEL_EXPORTER_JAEGER_ENDPOINT', 'http://loca
                        username: ENV['OTEL_EXPORTER_JAEGER_USER'],
                        password: ENV['OTEL_EXPORTER_JAEGER_PASSWORD'],
                        timeout: ENV.fetch('OTEL_EXPORTER_JAEGER_TIMEOUT', 10),
-                       ssl_verify_mode: CollectorExporter.ssl_verify_mode)
+                       ssl_verify_mode: CollectorExporter.ssl_verify_mode,
+                       metrics_reporter: nil)
           raise ArgumentError, "invalid url for Jaeger::CollectorExporter #{endpoint}" unless OpenTelemetry::Common::Utilities.valid_url?(endpoint)
           raise ArgumentError, 'username and password should either both be nil or both be set' if username.nil? != password.nil?
 
@@ -41,6 +42,7 @@ def initialize(endpoint: ENV.fetch('OTEL_EXPORTER_JAEGER_ENDPOINT', 'http://loca
             @transport.add_headers(auth_header)
           end
           @serializer = ::Thrift::Serializer.new
+          @metrics_reporter = metrics_reporter || OpenTelemetry::SDK::Trace::Export::MetricsReporter
           @shutdown = false
         end
 
@@ -59,10 +61,13 @@ def export(span_data, timeout: nil)
           end
 
           OpenTelemetry::Common::Utilities.untraced do
-            @transport.flush
+            measure_request_duration do
+              @transport.flush
+            end
           end
           SUCCESS
         rescue StandardError => e
+          @metrics_reporter.add_to_counter('otel.jaeger_exporter.failure', labels: { 'reason' => e.class.to_s })
           OpenTelemetry.handle_error(exception: e, message: 'unexpected error in Jaeger::CollectorExporter#export')
           FAILURE
         end
@@ -95,6 +100,15 @@ def encoded_batches(span_data)
             Thrift::Batch.new('process' => process, 'spans' => spans)
           end
         end
+
+        def measure_request_duration
+          start = Process.clock_gettime(Process::CLOCK_MONOTONIC)
+          yield
+        ensure
+          stop = Process.clock_gettime(Process::CLOCK_MONOTONIC)
+          duration_ms = 1000.0 * (stop - start)
+          @metrics_reporter.record_value('otel.jaeger_exporter.request_duration', value: duration_ms)
+        end
       end
     end
   end

exporter/jaeger/opentelemetry-exporter-jaeger.gemspec

@@ -37,6 +37,7 @@ Gem::Specification.new do |spec|
   spec.add_development_dependency 'minitest', '~> 5.0'
   spec.add_development_dependency 'opentelemetry-test-helpers'
   spec.add_development_dependency 'rake', '~> 12.0'
+  spec.add_development_dependency 'rspec-mocks'
   spec.add_development_dependency 'rubocop', '~> 0.73.0'
   spec.add_development_dependency 'simplecov', '~> 0.17'
   spec.add_development_dependency 'webmock', '~> 3.7.6'

exporter/jaeger/test/opentelemetry/exporters/jaeger/agent_exporter_test.rb renamed to exporter/jaeger/test/opentelemetry/exporter/jaeger/agent_exporter_test.rb

@@ -8,6 +8,11 @@
 DEFAULT_JAEGER_COLLECTOR_ENDPOINT = 'http://localhost:14268/api/traces'
 
 describe OpenTelemetry::Exporter::Jaeger::CollectorExporter do
+  let(:metrics_reporter) { double }
+  before(:each) do
+    allow(metrics_reporter).to receive(:record_value)
+  end
+
   describe '#initialize' do
     it 'initializes with defaults' do
       exp = OpenTelemetry::Exporter::Jaeger::CollectorExporter.new
@@ -175,5 +180,46 @@
       _(result).must_equal(OpenTelemetry::SDK::Trace::Export::SUCCESS)
       assert_requested(stub_post)
     end
+
+    describe '#metrics_reporter' do
+      it 'reports metrics for failures' do
+        expect(metrics_reporter).to receive(:add_to_counter).with('otel.jaeger_exporter.failure', labels: { 'reason' => 'Thrift::TransportException' })
+
+        stub_post = stub_request(:post, DEFAULT_JAEGER_COLLECTOR_ENDPOINT).to_return(status: 500)
+        exporter = OpenTelemetry::Exporter::Jaeger::CollectorExporter.new(
+          metrics_reporter: metrics_reporter
+        )
+        span_data = create_span_data
+        result = exporter.export([span_data])
+        _(result).must_equal(OpenTelemetry::SDK::Trace::Export::FAILURE)
+        assert_requested(stub_post)
+      end
+
+      it 'reports metrics for timeouts' do
+        expect(metrics_reporter).to receive(:add_to_counter).with('otel.jaeger_exporter.failure', labels: { 'reason' => 'Net::OpenTimeout' })
+
+        stub_post = stub_request(:post, DEFAULT_JAEGER_COLLECTOR_ENDPOINT).to_timeout
+        exporter = OpenTelemetry::Exporter::Jaeger::CollectorExporter.new(
+          metrics_reporter: metrics_reporter
+        )
+        span_data = create_span_data
+        result = exporter.export([span_data])
+        _(result).must_equal(OpenTelemetry::SDK::Trace::Export::FAILURE)
+        assert_requested(stub_post)
+      end
+
+      it 'records request duration' do
+        expect(metrics_reporter).to receive(:record_value).with('otel.jaeger_exporter.request_duration', hash_including(value: kind_of(Float)))
+        stub_post = stub_request(:post, DEFAULT_JAEGER_COLLECTOR_ENDPOINT).to_return(status: 200)
+
+        exporter = OpenTelemetry::Exporter::Jaeger::CollectorExporter.new(
+          metrics_reporter: metrics_reporter
+        )
+        span_data = create_span_data
+        result = exporter.export([span_data])
+        _(result).must_equal(OpenTelemetry::SDK::Trace::Export::SUCCESS)
+        assert_requested(stub_post)
+      end
+    end
   end
 end

exporter/jaeger/test/opentelemetry/exporters/jaeger/collector_exporter_test.rb renamed to exporter/jaeger/test/opentelemetry/exporter/jaeger/collector_exporter_test.rb

@@ -11,6 +11,7 @@
 require 'opentelemetry/exporter/jaeger'
 require 'minitest/autorun'
 require 'webmock/minitest'
+require 'rspec/mocks/minitest_integration'
 
 def create_span_data(name: '', kind: nil, status: nil, parent_span_id: OpenTelemetry::Trace::INVALID_SPAN_ID,
                      total_recorded_attributes: 0, total_recorded_events: 0, total_recorded_links: 0, start_timestamp: OpenTelemetry::TestHelpers.exportable_timestamp,