improve explain telemetry

hasura · Feb 9, 2024 · 971515f · 971515f
1 parent 848a5bc
commit 971515f
Show file tree

Hide file tree

Showing 2 changed files with 38 additions and 23 deletions.
diff --git a/connector/server.go b/connector/server.go
@@ -261,7 +261,7 @@ func (s *Server[RawConfiguration, Configuration, State]) QueryExplain(w http.Res
 			attribute.String("reason", "json_decode"),
 		}
 		span.SetAttributes(attributes...)
-		s.telemetry.explainCounter.Add(r.Context(), 1, metric.WithAttributes(attributes...))
+		s.telemetry.queryExplainCounter.Add(r.Context(), 1, metric.WithAttributes(attributes...))
 		return
 	}
 	decodeSpan.End()
@@ -280,7 +280,7 @@ func (s *Server[RawConfiguration, Configuration, State]) QueryExplain(w http.Res
 			attribute.String("reason", fmt.Sprintf("%d", status)),
 		}
 		span.SetAttributes(attributes...)
-		s.telemetry.explainCounter.Add(r.Context(), 1, metric.WithAttributes(append(attributes, statusAttributes...)...))
+		s.telemetry.queryExplainCounter.Add(r.Context(), 1, metric.WithAttributes(append(attributes, statusAttributes...)...))
 		return
 	}
 	execSpan.End()
@@ -290,10 +290,10 @@ func (s *Server[RawConfiguration, Configuration, State]) QueryExplain(w http.Res
 	_, responseSpan := s.telemetry.Tracer.Start(ctx, "Response")
 	writeJson(w, logger, http.StatusOK, response)
 	responseSpan.End()
-	s.telemetry.explainCounter.Add(r.Context(), 1, metric.WithAttributes(append(attributes, statusAttribute)...))
+	s.telemetry.queryExplainCounter.Add(r.Context(), 1, metric.WithAttributes(append(attributes, statusAttribute)...))
 
 	// record latency for success requests only
-	s.telemetry.explainLatencyHistogram.Record(r.Context(), time.Since(startTime).Seconds(), metric.WithAttributes(collectionAttr))
+	s.telemetry.queryExplainLatencyHistogram.Record(r.Context(), time.Since(startTime).Seconds(), metric.WithAttributes(collectionAttr))
 }
 
 // MutationExplain implements a handler for the /mutation/explain endpoint, POST method that explains a mutation by creating an execution plan.
@@ -320,7 +320,7 @@ func (s *Server[RawConfiguration, Configuration, State]) MutationExplain(w http.
 			attribute.String("reason", "json_decode"),
 		}
 		span.SetAttributes(attributes...)
-		s.telemetry.explainCounter.Add(r.Context(), 1, metric.WithAttributes(attributes...))
+		s.telemetry.mutationExplainCounter.Add(r.Context(), 1, metric.WithAttributes(attributes...))
 		return
 	}
 	decodeSpan.End()
@@ -343,7 +343,7 @@ func (s *Server[RawConfiguration, Configuration, State]) MutationExplain(w http.
 			attribute.String("reason", fmt.Sprintf("%d", status)),
 		}
 		span.SetAttributes(attributes...)
-		s.telemetry.explainCounter.Add(r.Context(), 1, metric.WithAttributes(append(attributes, statusAttributes...)...))
+		s.telemetry.mutationExplainCounter.Add(r.Context(), 1, metric.WithAttributes(append(attributes, statusAttributes...)...))
 		return
 	}
 	execSpan.End()
@@ -353,10 +353,10 @@ func (s *Server[RawConfiguration, Configuration, State]) MutationExplain(w http.
 	_, responseSpan := s.telemetry.Tracer.Start(ctx, "Response")
 	writeJson(w, logger, http.StatusOK, response)
 	responseSpan.End()
-	s.telemetry.explainCounter.Add(r.Context(), 1, metric.WithAttributes(append(attributes, statusAttribute)...))
+	s.telemetry.mutationExplainCounter.Add(r.Context(), 1, metric.WithAttributes(append(attributes, statusAttribute)...))
 
 	// record latency for success requests only
-	s.telemetry.explainLatencyHistogram.Record(r.Context(), time.Since(startTime).Seconds(), metric.WithAttributes(collectionAttr))
+	s.telemetry.mutationExplainLatencyHistogram.Record(r.Context(), time.Since(startTime).Seconds(), metric.WithAttributes(collectionAttr))
 }
 
 // Mutation implements a handler for the /mutation endpoint, POST method that executes a mutation.

diff --git a/connector/telemetry.go b/connector/telemetry.go
@@ -27,15 +27,17 @@ import (
 )
 
 type TelemetryState struct {
-	Tracer                   traceapi.Tracer
-	Meter                    metricapi.Meter
-	Shutdown                 func(context.Context) error
-	queryCounter             metricapi.Int64Counter
-	mutationCounter          metricapi.Int64Counter
-	explainCounter           metricapi.Int64Counter
-	queryLatencyHistogram    metricapi.Float64Histogram
-	explainLatencyHistogram  metricapi.Float64Histogram
-	mutationLatencyHistogram metricapi.Float64Histogram
+	Tracer                          traceapi.Tracer
+	Meter                           metricapi.Meter
+	Shutdown                        func(context.Context) error
+	queryCounter                    metricapi.Int64Counter
+	mutationCounter                 metricapi.Int64Counter
+	queryExplainCounter             metricapi.Int64Counter
+	mutationExplainCounter          metricapi.Int64Counter
+	queryLatencyHistogram           metricapi.Float64Histogram
+	queryExplainLatencyHistogram    metricapi.Float64Histogram
+	mutationExplainLatencyHistogram metricapi.Float64Histogram
+	mutationLatencyHistogram        metricapi.Float64Histogram
 }
 
 // setupOTelSDK bootstraps the OpenTelemetry pipeline.
@@ -226,9 +228,17 @@ func setupMetrics(telemetry *TelemetryState, metricsPrefix string) error {
 		return err
 	}
 
-	telemetry.explainCounter, err = meter.Int64Counter(
-		fmt.Sprintf("%sexplain.total", metricsPrefix),
-		metricapi.WithDescription("The total number of explain requests"),
+	telemetry.queryExplainCounter, err = meter.Int64Counter(
+		fmt.Sprintf("%squery.explain_total", metricsPrefix),
+		metricapi.WithDescription("The total number of explain query requests"),
+	)
+	if err != nil {
+		return err
+	}
+
+	telemetry.mutationExplainCounter, err = meter.Int64Counter(
+		fmt.Sprintf("%smutation.explain_total", metricsPrefix),
+		metricapi.WithDescription("The total number of explain mutation requests"),
 	)
 	if err != nil {
 		return err
@@ -250,9 +260,14 @@ func setupMetrics(telemetry *TelemetryState, metricsPrefix string) error {
 		return err
 	}
 
-	telemetry.explainLatencyHistogram, err = meter.Float64Histogram(
-		fmt.Sprintf("%sexplain.total_time", metricsPrefix),
-		metricapi.WithDescription("Total time taken to plan and execute an explain request, in seconds"),
+	telemetry.queryExplainLatencyHistogram, err = meter.Float64Histogram(
+		fmt.Sprintf("%squery.explain_total_time", metricsPrefix),
+		metricapi.WithDescription("Total time taken to plan and execute an explain query request, in seconds"),
+	)
+
+	telemetry.mutationExplainLatencyHistogram, err = meter.Float64Histogram(
+		fmt.Sprintf("%smutation.explain_total_time", metricsPrefix),
+		metricapi.WithDescription("Total time taken to plan and execute an explain mutation request, in seconds"),
 	)
 
 	return err