refactor: unify error metrics (#1342)

wundergraph · Nov 7, 2024 · c6277bd · c6277bd
1 parent fb47157
commit c6277bd
Show file tree

Hide file tree

Showing 8 changed files with 205 additions and 114 deletions.
diff --git a/router-tests/prometheus_test.go b/router-tests/prometheus_test.go
@@ -1645,8 +1645,8 @@ func TestPrometheus(t *testing.T) {
 			totalRequestErrorsMetric := totalRequestsErrors.GetMetric()
 
 			require.Len(t, totalRequestErrorsMetric, 2)
-			require.Len(t, totalRequestErrorsMetric[0].Label, 13)
-			require.Len(t, totalRequestErrorsMetric[1].Label, 15)
+			require.Len(t, totalRequestErrorsMetric[0].Label, 12)
+			require.Len(t, totalRequestErrorsMetric[1].Label, 14)
 
 			// Error metric for the subgraph error
 			require.Equal(t, []*io_prometheus_client.LabelPair{
@@ -1686,10 +1686,6 @@ func TestPrometheus(t *testing.T) {
 					Name:  PointerOf("wg_operation_type"),
 					Value: PointerOf("query"),
 				},
-				{
-					Name:  PointerOf("wg_request_error"),
-					Value: PointerOf("true"),
-				},
 				{
 					Name:  PointerOf("wg_router_cluster_name"),
 					Value: PointerOf(""),
@@ -1726,10 +1722,6 @@ func TestPrometheus(t *testing.T) {
 					Name:  PointerOf("wg_client_version"),
 					Value: PointerOf("missing"),
 				},
-				{
-					Name:  PointerOf("wg_component_name"),
-					Value: PointerOf("engine-loader"),
-				},
 				{
 					Name:  PointerOf("wg_federated_graph_id"),
 					Value: PointerOf("graph"),
@@ -1807,13 +1799,39 @@ func TestPrometheus(t *testing.T) {
 			mf, err := promRegistry.Gather()
 			require.NoError(t, err)
 
+			requestDuration := findMetricFamilyByName(mf, "router_http_request_duration_milliseconds")
+			requestDurationMetric := requestDuration.GetMetric()
+
+			/**
+			employees -> 200 Status code = 1
+			products -> 402 + 2x error codes = 2
+			router -> 200 Status code + 2x error codes = 2
+
+			Total metrics = 5
+			*/
+
+			require.Len(t, requestDurationMetric, 5)
+			require.Len(t, requestDurationMetric[0].Label, 14)
+			require.Len(t, requestDurationMetric[1].Label, 14)
+			require.Len(t, requestDurationMetric[2].Label, 14)
+			require.Len(t, requestDurationMetric[3].Label, 16)
+			require.Len(t, requestDurationMetric[4].Label, 16)
+
 			totalRequestsErrors := findMetricFamilyByName(mf, "router_http_requests_error_total")
 			totalRequestErrorsMetric := totalRequestsErrors.GetMetric()
 
-			require.Len(t, totalRequestErrorsMetric, 3)
-			require.Len(t, totalRequestErrorsMetric[0].Label, 14)
-			require.Len(t, totalRequestErrorsMetric[1].Label, 14)
+			/**
+			products -> 402 + 2x error codes = 2
+			router -> 200 Status code + 2x error codes = 2
+
+			Total metrics = 4
+			*/
+
+			require.Len(t, totalRequestErrorsMetric, 4)
+			require.Len(t, totalRequestErrorsMetric[0].Label, 13)
+			require.Len(t, totalRequestErrorsMetric[1].Label, 13)
 			require.Len(t, totalRequestErrorsMetric[2].Label, 15)
+			require.Len(t, totalRequestErrorsMetric[3].Label, 15)
 
 			require.Equal(t, []*io_prometheus_client.LabelPair{
 				{
@@ -1856,10 +1874,6 @@ func TestPrometheus(t *testing.T) {
 					Name:  PointerOf("wg_operation_type"),
 					Value: PointerOf("query"),
 				},
-				{
-					Name:  PointerOf("wg_request_error"),
-					Value: PointerOf("true"),
-				},
 				{
 					Name:  PointerOf("wg_router_cluster_name"),
 					Value: PointerOf(""),
@@ -1916,10 +1930,6 @@ func TestPrometheus(t *testing.T) {
 					Name:  PointerOf("wg_operation_type"),
 					Value: PointerOf("query"),
 				},
-				{
-					Name:  PointerOf("wg_request_error"),
-					Value: PointerOf("true"),
-				},
 				{
 					Name:  PointerOf("wg_router_cluster_name"),
 					Value: PointerOf(""),
@@ -1936,6 +1946,10 @@ func TestPrometheus(t *testing.T) {
 
 			// Error metric for the subgraph error
 			require.Equal(t, []*io_prometheus_client.LabelPair{
+				{
+					Name:  PointerOf("error_codes"),
+					Value: PointerOf("UNAUTHORIZED"),
+				},
 				{
 					Name:  PointerOf("http_status_code"),
 					Value: PointerOf("403"),
@@ -1957,8 +1971,67 @@ func TestPrometheus(t *testing.T) {
 					Value: PointerOf("missing"),
 				},
 				{
-					Name:  PointerOf("wg_component_name"),
-					Value: PointerOf("engine-loader"),
+					Name:  PointerOf("wg_federated_graph_id"),
+					Value: PointerOf("graph"),
+				},
+				{
+					Name:  PointerOf("wg_operation_name"),
+					Value: PointerOf("myQuery"),
+				},
+				{
+					Name:  PointerOf("wg_operation_protocol"),
+					Value: PointerOf("http"),
+				},
+				{
+					Name:  PointerOf("wg_operation_type"),
+					Value: PointerOf("query"),
+				},
+				{
+					Name:  PointerOf("wg_router_cluster_name"),
+					Value: PointerOf(""),
+				},
+				{
+					Name:  PointerOf("wg_router_config_version"),
+					Value: PointerOf(xEnv.RouterConfigVersionMain()),
+				},
+				{
+					Name:  PointerOf("wg_router_version"),
+					Value: PointerOf("dev"),
+				},
+				{
+					Name:  PointerOf("wg_subgraph_id"),
+					Value: PointerOf("3"),
+				},
+				{
+					Name:  PointerOf("wg_subgraph_name"),
+					Value: PointerOf("products"),
+				},
+			}, totalRequestErrorsMetric[2].Label)
+
+			require.Equal(t, []*io_prometheus_client.LabelPair{
+				{
+					Name:  PointerOf("error_codes"),
+					Value: PointerOf("YOUR_ERROR_CODE"),
+				},
+				{
+					Name:  PointerOf("http_status_code"),
+					Value: PointerOf("403"),
+				},
+				{
+					Name:  PointerOf("otel_scope_name"),
+					Value: PointerOf("cosmo.router.prometheus"),
+				},
+				{
+					Name:  PointerOf("otel_scope_version"),
+					Value: PointerOf("0.0.1"),
+				},
+				{
+					Name:  PointerOf("wg_client_name"),
+					Value: PointerOf("unknown"),
+				},
+				{
+					Name:  PointerOf("wg_client_version"),
+					Value: PointerOf("missing"),
 				},
 				{
 					Name:  PointerOf("wg_federated_graph_id"),
@@ -1996,7 +2069,7 @@ func TestPrometheus(t *testing.T) {
 					Name:  PointerOf("wg_subgraph_name"),
 					Value: PointerOf("products"),
 				},
-			}, totalRequestErrorsMetric[2].Label)
+			}, totalRequestErrorsMetric[3].Label)
 		})
 	})
 

diff --git a/router-tests/telemetry_test.go b/router-tests/telemetry_test.go
@@ -2349,16 +2349,16 @@ func TestTelemetry(t *testing.T) {
 			require.Equal(t, codes.Error, sn[1].Status().Code)
 			require.Contains(t, sn[1].Status().Description, "unexpected literal - got: UNDEFINED want one of: [ENUM TYPE UNION QUERY INPUT EXTEND SCHEMA SCALAR FRAGMENT INTERFACE DIRECTIVE]")
 
-			require.Lenf(t, sn[1].Attributes(), 8, "expected 14 attributes, got %d", len(sn[1].Attributes()))
+			require.Lenf(t, sn[1].Attributes(), 8, "expected 8 attributes, got %d", len(sn[1].Attributes()))
 
 			require.Contains(t, sn[1].Attributes(), otel.WgRouterVersion.String("dev"))
 			require.Contains(t, sn[1].Attributes(), otel.WgRouterClusterName.String(""))
 			require.Contains(t, sn[1].Attributes(), otel.WgFederatedGraphID.String("graph"))
 			require.Contains(t, sn[1].Attributes(), otel.WgRouterConfigVersion.String(xEnv.RouterConfigVersionMain()))
 			require.Contains(t, sn[1].Attributes(), otel.WgClientName.String("unknown"))
+			require.Contains(t, sn[1].Attributes(), otel.WgRequestError.Bool(true))
 			require.Contains(t, sn[1].Attributes(), otel.WgClientVersion.String("missing"))
 			require.Contains(t, sn[1].Attributes(), otel.WgOperationProtocol.String("http"))
-			require.Contains(t, sn[1].Attributes(), otel.WgRequestError.Bool(true))
 
 			events := sn[1].Events()
 			require.Len(t, events, 1, "expected 1 event because the GraphQL parsing failed")
@@ -2370,7 +2370,6 @@ func TestTelemetry(t *testing.T) {
 			require.Contains(t, sn[2].Status().Description, "unexpected literal - got: UNDEFINED want one of: [ENUM TYPE UNION QUERY INPUT EXTEND SCHEMA SCALAR FRAGMENT INTERFACE DIRECTIVE]")
 
 			require.Lenf(t, sn[2].Attributes(), 23, "expected 23 attributes, got %d", len(sn[2].Attributes()))
-			require.Contains(t, sn[2].Attributes(), otel.WgRequestError.Bool(true))
 
 			events = sn[2].Events()
 			require.Len(t, events, 1, "expected 1 event because the GraphQL request failed")
@@ -2745,8 +2744,10 @@ func TestTelemetry(t *testing.T) {
 							{
 								Attributes: attribute.NewSet(
 									attribute.String("from_header", "custom-value"),
+									attribute.StringSlice("error_codes", []string{"UNAUTHORIZED", "YOUR_ERROR_CODE"}),
 									attribute.String("sha256", "b0066f89f91315b4610ed127be677e6cea380494eb20c83cc121c97552ca44b2"),
 									semconv.HTTPStatusCode(403),
+									otel.WgRequestError.Bool(true),
 									otel.WgClientName.String("unknown"),
 									otel.WgClientVersion.String("missing"),
 									otel.WgFederatedGraphID.String("graph"),
@@ -2818,7 +2819,9 @@ func TestTelemetry(t *testing.T) {
 							{
 								Attributes: attribute.NewSet(
 									attribute.String("from_header", "custom-value"),
+									attribute.StringSlice("error_codes", []string{"UNAUTHORIZED", "YOUR_ERROR_CODE"}),
 									attribute.String("sha256", "b0066f89f91315b4610ed127be677e6cea380494eb20c83cc121c97552ca44b2"),
+									otel.WgRequestError.Bool(true),
 									semconv.HTTPStatusCode(403),
 									otel.WgClientName.String("unknown"),
 									otel.WgClientVersion.String("missing"),
@@ -2837,12 +2840,12 @@ func TestTelemetry(t *testing.T) {
 							},
 							{
 								Attributes: attribute.NewSet(
-									otel.WgRequestError.Bool(true),
 									attribute.StringSlice("error_codes", []string{"UNAUTHORIZED", "YOUR_ERROR_CODE"}),
 									attribute.StringSlice("error_services", []string{"products"}),
 									attribute.String("from_header", "custom-value"),
 									semconv.HTTPStatusCode(200),
 									attribute.StringSlice("services", []string{"employees", "products"}),
+									otel.WgRequestError.Bool(true),
 									attribute.String("sha256", "b0066f89f91315b4610ed127be677e6cea380494eb20c83cc121c97552ca44b2"),
 									otel.WgClientName.String("unknown"),
 									otel.WgClientVersion.String("missing"),
@@ -2916,7 +2919,6 @@ func TestTelemetry(t *testing.T) {
 									attribute.StringSlice("services", []string{"employees", "products"}),
 									attribute.StringSlice("error_services", []string{"products"}),
 									semconv.HTTPStatusCode(200),
-									otel.WgRequestError.Bool(true),
 									otel.WgClientName.String("unknown"),
 									otel.WgClientVersion.String("missing"),
 									otel.WgFederatedGraphID.String("graph"),
@@ -3009,7 +3011,6 @@ func TestTelemetry(t *testing.T) {
 									attribute.StringSlice("services", []string{"employees", "products"}),
 									attribute.StringSlice("error_services", []string{"products"}),
 									semconv.HTTPStatusCode(200),
-									otel.WgRequestError.Bool(true),
 									otel.WgClientName.String("unknown"),
 									otel.WgClientVersion.String("missing"),
 									otel.WgFederatedGraphID.String("graph"),
@@ -3129,9 +3130,9 @@ func TestTelemetry(t *testing.T) {
 							{
 								Attributes: attribute.NewSet(
 									attribute.String("from_header", "custom-value"),
+									attribute.StringSlice("error_codes", []string{"UNAUTHORIZED", "YOUR_ERROR_CODE"}),
 									attribute.String("sha256", "b0066f89f91315b4610ed127be677e6cea380494eb20c83cc121c97552ca44b2"),
 									semconv.HTTPStatusCode(403),
-									otel.WgComponentName.String("engine-loader"),
 									otel.WgClientName.String("unknown"),
 									otel.WgClientVersion.String("missing"),
 									otel.WgFederatedGraphID.String("graph"),
@@ -3165,7 +3166,6 @@ func TestTelemetry(t *testing.T) {
 									otel.WgRouterClusterName.String(""),
 									otel.WgRouterConfigVersion.String(xEnv.RouterConfigVersionMain()),
 									otel.WgRouterVersion.String("dev"),
-									otel.WgRequestError.Bool(true),
 								),
 								Value: 1,
 							},
@@ -3189,6 +3189,7 @@ func TestTelemetry(t *testing.T) {
 						failedRequestsMetric,
 					},
 				}
+
 				metricdatatest.AssertEqual(t, want, rm.ScopeMetrics[0], metricdatatest.IgnoreTimestamp(), metricdatatest.IgnoreValue())
 			})
 		})

diff --git a/router/core/context.go b/router/core/context.go
@@ -27,6 +27,7 @@ type contextKey int
 const (
 	requestContextKey contextKey = iota
 	subgraphResolverContextKey
+	engineLoaderHooksContextKey
 )
 
 var _ RequestContext = (*requestContext)(nil)