fix: metric reset not effect

Signed-off-by: nayihz <[email protected]>

Author: nayihz

cmd/bbr/main.go

@@ -32,8 +32,8 @@ import (
 	metricsserver "sigs.k8s.io/controller-runtime/pkg/metrics/server"
 
 	"sigs.k8s.io/gateway-api-inference-extension/internal/runnable"
+	"sigs.k8s.io/gateway-api-inference-extension/pkg/bbr/metrics"
 	runserver "sigs.k8s.io/gateway-api-inference-extension/pkg/bbr/server"
-	"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/metrics"
 	"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/util/logging"
 )
 

cmd/epp/main.go

@@ -154,10 +154,6 @@ func run() error {
 		return err
 	}
 
-	poolNamespacedName := types.NamespacedName{
-		Name:      *poolName,
-		Namespace: *poolNamespace,
-	}
 	metrics.Register()
 	// Register metrics handler.
 	// Metrics endpoint is enabled in 'config/default/kustomization.yaml'. The Metrics options configure the server.
@@ -169,6 +165,10 @@ func run() error {
 		FilterProvider: filters.WithAuthenticationAndAuthorization,
 	}
 
+	poolNamespacedName := types.NamespacedName{
+		Name:      *poolName,
+		Namespace: *poolNamespace,
+	}
 	mgr, err := runserver.NewDefaultManager(poolNamespacedName, cfg, metricsServerOptions)
 	if err != nil {
 		setupLog.Error(err, "Failed to create controller manager")

pkg/epp/metrics/metrics.go

@@ -309,20 +309,20 @@ func Register() {
 }
 
 // Just for integration test
-func Unregister() {
-	metrics.Registry.Unregister(requestCounter)
-	metrics.Registry.Unregister(requestErrCounter)
-	metrics.Registry.Unregister(requestLatencies)
-	metrics.Registry.Unregister(requestSizes)
-	metrics.Registry.Unregister(responseSizes)
-	metrics.Registry.Unregister(inputTokens)
-	metrics.Registry.Unregister(outputTokens)
-	metrics.Registry.Unregister(runningRequests)
-	metrics.Registry.Unregister(NormalizedTimePerOutputToken)
-
-	metrics.Registry.Unregister(inferencePoolAvgKVCache)
-	metrics.Registry.Unregister(inferencePoolAvgQueueSize)
-	metrics.Registry.Unregister(inferencePoolReadyPods)
+func Reset() {
+	requestCounter.Reset()
+	requestErrCounter.Reset()
+	requestLatencies.Reset()
+	requestSizes.Reset()
+	responseSizes.Reset()
+	inputTokens.Reset()
+	outputTokens.Reset()
+	runningRequests.Reset()
+	NormalizedTimePerOutputToken.Reset()
+	inferencePoolAvgKVCache.Reset()
+	inferencePoolAvgQueueSize.Reset()
+	inferencePoolReadyPods.Reset()
+	SchedulerPluginProcessingLatencies.Reset()
 }
 
 // RecordRequstCounter records the number of requests.

pkg/epp/metrics/testdata/scheduler_e2e_duration_seconds_metric

@@ -1,4 +1,4 @@
-# HELP inference_extension_scheduler_e2e_duration_seconds [ALPHA] End-to-end scheduling latency distribution in seconds.
+# HELP inference_extension_scheduler_e2e_duration_seconds End-to-end scheduling latency distribution in seconds.
 # TYPE inference_extension_scheduler_e2e_duration_seconds histogram
 inference_extension_scheduler_e2e_duration_seconds_bucket{le="0.0001"} 0
 inference_extension_scheduler_e2e_duration_seconds_bucket{le="0.0002"} 1

pkg/epp/metrics/testdata/scheduler_plugin_processing_latencies_metric

@@ -1,4 +1,4 @@
-# HELP inference_extension_scheduler_plugin_duration_seconds [ALPHA] Scheduler plugin processing latency distribution in seconds for each plugin type and plugin name.
+# HELP inference_extension_scheduler_plugin_duration_seconds Scheduler plugin processing latency distribution in seconds for each plugin type and plugin name.
 # TYPE inference_extension_scheduler_plugin_duration_seconds histogram
 inference_extension_scheduler_plugin_duration_seconds_bucket{plugin_name="PluginA",plugin_type="PreSchedule",le="0.0001"} 0
 inference_extension_scheduler_plugin_duration_seconds_bucket{plugin_name="PluginA",plugin_type="PreSchedule",le="0.0002"} 0

test/integration/epp/hermetic_test.go

@@ -1258,7 +1258,6 @@ func TestFullDuplexStreamed_KubeInferenceModelRequest(t *testing.T) {
 
 	for _, test := range tests {
 		t.Run(test.name, func(t *testing.T) {
-			metrics.Register()
 			client, cleanup := setUpHermeticServer(t, test.pods, true)
 			t.Cleanup(cleanup)
 			responses, err := integrationutils.StreamedRequest(t, client, test.requests, len(test.wantResponses))
@@ -1277,9 +1276,8 @@ func TestFullDuplexStreamed_KubeInferenceModelRequest(t *testing.T) {
 					}
 				}
 			}
-			metrics.Unregister()
+			metrics.Reset()
 		})
-		return
 	}
 }
 
@@ -1394,6 +1392,7 @@ func BeforeSuite() func() {
 	// Init runtime.
 	ctrl.SetLogger(logger)
 
+	metrics.Register()
 	// Register metrics handler.
 	// Metrics endpoint is enabled in 'config/default/kustomization.yaml'. The Metrics options configure the server.
 	// More info: