exporter.rb in opentelemetry-exporter-otlp-0.11.0

- old
+ new

@@ -29,16 +29,17 @@
         KEEP_ALIVE_TIMEOUT = 30
         RETRY_COUNT = 5
         WRITE_TIMEOUT_SUPPORTED = Gem::Version.new(RUBY_VERSION) >= Gem::Version.new('2.6')
         private_constant(:KEEP_ALIVE_TIMEOUT, :RETRY_COUNT, :WRITE_TIMEOUT_SUPPORTED)
 
-        def initialize(endpoint: config_opt('OTEL_EXPORTER_OTLP_SPAN_ENDPOINT', 'OTEL_EXPORTER_OTLP_ENDPOINT', default: 'localhost:55681/v1/trace'), # rubocop:disable Metrics/AbcSize, Metrics/CyclomaticComplexity
+        def initialize(endpoint: config_opt('OTEL_EXPORTER_OTLP_SPAN_ENDPOINT', 'OTEL_EXPORTER_OTLP_ENDPOINT', default: 'localhost:55681/v1/trace'), # rubocop:disable Metrics/AbcSize, Metrics/CyclomaticComplexity, Metrics/PerceivedComplexity
                        insecure: config_opt('OTEL_EXPORTER_OTLP_SPAN_INSECURE', 'OTEL_EXPORTER_OTLP_INSECURE', default: false),
                        certificate_file: config_opt('OTEL_EXPORTER_OTLP_SPAN_CERTIFICATE', 'OTEL_EXPORTER_OTLP_CERTIFICATE'),
                        headers: config_opt('OTEL_EXPORTER_OTLP_SPAN_HEADERS', 'OTEL_EXPORTER_OTLP_HEADERS'), # TODO: what format is expected here?
                        compression: config_opt('OTEL_EXPORTER_OTLP_SPAN_COMPRESSION', 'OTEL_EXPORTER_OTLP_COMPRESSION'),
-                       timeout: config_opt('OTEL_EXPORTER_OTLP_SPAN_TIMEOUT', 'OTEL_EXPORTER_OTLP_TIMEOUT', default: 10))
+                       timeout: config_opt('OTEL_EXPORTER_OTLP_SPAN_TIMEOUT', 'OTEL_EXPORTER_OTLP_TIMEOUT', default: 10),
+                       metrics_reporter: nil)
           raise ArgumentError, "invalid url for OTLP::Exporter #{endpoint}" if invalid_url?("http://#{endpoint}")
           raise ArgumentError, "unsupported compression key #{compression}" unless compression.nil? || compression == 'gzip'
           raise ArgumentError, 'headers must be comma-separated k:v pairs or a Hash' unless valid_headers?(headers)
 
           uri = URI "http://#{endpoint}"
@@ -52,10 +53,11 @@
                      when String then CSV.parse(headers, col_sep: ':', row_sep: ',').to_h
                      when Hash then headers
                      end
           @timeout = timeout.to_f # TODO: use this as a default timeout when we implement timeouts in https://github.com/open-telemetry/opentelemetry-ruby/pull/341
           @compression = compression
+          @metrics_reporter = metrics_reporter || OpenTelemetry::SDK::Trace::Export::MetricsReporter
 
           @shutdown = false
         end
 
         # Called to export sampled {OpenTelemetry::SDK::Trace::SpanData} structs.
@@ -130,38 +132,38 @@
 
             @http.open_timeout = remaining_timeout
             @http.read_timeout = remaining_timeout
             @http.write_timeout = remaining_timeout if WRITE_TIMEOUT_SUPPORTED
             @http.start unless @http.started?
-            response = @http.request(request)
+            response = measure_request_duration { @http.request(request) }
 
             case response
             when Net::HTTPOK
               response.body # Read and discard body
               SUCCESS
             when Net::HTTPServiceUnavailable, Net::HTTPTooManyRequests
               response.body # Read and discard body
-              redo if backoff?(retry_after: response['Retry-After'], retry_count: retry_count += 1)
+              redo if backoff?(retry_after: response['Retry-After'], retry_count: retry_count += 1, reason: response.code)
               FAILURE
             when Net::HTTPRequestTimeOut, Net::HTTPGatewayTimeOut, Net::HTTPBadGateway
               response.body # Read and discard body
-              redo if backoff?(retry_count: retry_count += 1)
+              redo if backoff?(retry_count: retry_count += 1, reason: response.code)
               FAILURE
             when Net::HTTPBadRequest, Net::HTTPClientError, Net::HTTPServerError
               # TODO: decode the body as a google.rpc.Status Protobuf-encoded message when https://github.com/open-telemetry/opentelemetry-collector/issues/1357 is fixed.
               response.body # Read and discard body
               FAILURE
             when Net::HTTPRedirection
               @http.finish
               handle_redirect(response['location'])
-              redo if backoff?(retry_after: 0, retry_count: retry_count += 1)
+              redo if backoff?(retry_after: 0, retry_count: retry_count += 1, reason: response.code)
             else
               @http.finish
               FAILURE
             end
           rescue Net::OpenTimeout, Net::ReadTimeout
-            retry if backoff?(retry_count: retry_count += 1)
+            retry if backoff?(retry_count: retry_count += 1, reason: 'timeout')
             return FAILURE
           end
         ensure
           # Reset timeouts to defaults for the next call.
           @http.open_timeout = @timeout
@@ -175,13 +177,28 @@
 
         def untraced
           OpenTelemetry::Trace.with_span(OpenTelemetry::Trace::Span.new) { yield }
         end
 
-        def backoff?(retry_after: nil, retry_count:)
+        def measure_request_duration
+          start = Process.clock_gettime(Process::CLOCK_MONOTONIC)
+          begin
+            response = yield
+          ensure
+            stop = Process.clock_gettime(Process::CLOCK_MONOTONIC)
+            duration_ms = 1000.0 * (stop - start)
+            @metrics_reporter.record_value('otel.otlp_exporter.request_duration',
+                                           value: duration_ms,
+                                           labels: { 'status' => response&.code || 'unknown' })
+          end
+        end
+
+        def backoff?(retry_after: nil, retry_count:, reason:)
           return false if retry_count > RETRY_COUNT
 
+          @metrics_reporter.add_to_counter('otel.otlp_exporter.failure', labels: { 'reason' => reason })
+
           sleep_interval = nil
           unless retry_after.nil?
             sleep_interval =
               begin
                 Integer(retry_after)
@@ -231,11 +248,11 @@
 
         def as_otlp_span(span_data) # rubocop:disable Metrics/AbcSize, Metrics/MethodLength
           Opentelemetry::Proto::Trace::V1::Span.new(
             trace_id: span_data.trace_id,
             span_id: span_data.span_id,
-            trace_state: span_data.tracestate,
+            trace_state: span_data.tracestate.to_s,
             parent_span_id: span_data.parent_span_id == OpenTelemetry::Trace::INVALID_SPAN_ID ? nil : span_data.parent_span_id,
             name: span_data.name,
             kind: as_otlp_span_kind(span_data.kind),
             start_time_unix_nano: as_otlp_timestamp(span_data.start_timestamp),
             end_time_unix_nano: as_otlp_timestamp(span_data.end_timestamp),
@@ -252,10 +269,10 @@
             dropped_events_count: span_data.total_recorded_events - span_data.events&.size.to_i,
             links: span_data.links&.map do |link|
               Opentelemetry::Proto::Trace::V1::Span::Link.new(
                 trace_id: link.span_context.trace_id,
                 span_id: link.span_context.span_id,
-                trace_state: link.span_context.tracestate,
+                trace_state: link.span_context.tracestate.to_s,
                 attributes: link.attributes&.map { |k, v| as_otlp_key_value(k, v) }
                 # TODO: track dropped_attributes_count in Span#trim_links
               )
             end,
             dropped_links_count: span_data.total_recorded_links - span_data.links&.size.to_i,