Merge pull request #137466 from richabanker/nh-scheduler

Enable native histograms gated by feature flag in scheduler

cmd/kube-scheduler/app/options/options.go

@@ -47,6 +47,7 @@ import (
 	"k8s.io/component-base/logs"
 	logsapi "k8s.io/component-base/logs/api/v1"
 	"k8s.io/component-base/metrics"
+	metricsfeatures "k8s.io/component-base/metrics/features"
 	zpagesfeatures "k8s.io/component-base/zpages/features"
 	"k8s.io/klog/v2"
 	schedulerappconfig "k8s.io/kubernetes/cmd/kube-scheduler/app/config"
@@ -261,6 +262,7 @@ func (o *Options) ApplyTo(logger klog.Logger, c *schedulerappconfig.Config) erro
 			return err
 		}
 	}
+	metricsfeatures.ApplyFeatureGates(utilfeature.DefaultFeatureGate)
 	o.Metrics.Apply()
 
 	// Apply value independently instead of using ApplyDeprecated() because it can't be configured via ComponentConfig.

test/integration/metrics/nativehistograms/metrics_test.go

@@ -18,8 +18,13 @@ package nativehistograms
 
 import (
 	"context"
+	"crypto/tls"
+	"crypto/x509"
 	"fmt"
 	"net/http"
+	"os"
+	"path"
+	"strings"
 	"testing"
 
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -30,7 +35,9 @@ import (
 	metricsfeatures "k8s.io/component-base/metrics/features"
 	"k8s.io/component-base/metrics/testutil"
 	kubeapiservertesting "k8s.io/kubernetes/cmd/kube-apiserver/app/testing"
+	kubeschedulertesting "k8s.io/kubernetes/cmd/kube-scheduler/app/testing"
 	"k8s.io/kubernetes/test/integration/framework"
+	"k8s.io/kubernetes/test/utils/ktesting"
 )
 
 func scrapeMetrics(s *kubeapiservertesting.TestServer) (testutil.Metrics, error) {
@@ -101,3 +108,88 @@ func TestAPIServerNativeHistogramMetrics(t *testing.T) {
 		"apiserver_request_duration_seconds_count",
 	})
 }
+
+// TestSchedulerNativeHistogramMetrics verifies that native histogram metrics are properly
+// exposed in scheduler when the NativeHistograms feature gate is enabled.
+func TestSchedulerNativeHistogramMetrics(t *testing.T) {
+	featuregatetesting.SetFeatureGateDuringTest(t, utilfeature.DefaultFeatureGate, metricsfeatures.NativeHistograms, true)
+	s := kubeapiservertesting.StartTestServerOrDie(t, nil, framework.DefaultTestServerFlags(), framework.SharedEtcd())
+	defer s.TearDownFn()
+
+	apiserverConfig, err := os.CreateTemp("", "kubeconfig")
+	if err != nil {
+		t.Fatalf("Failed to create config file: %v", err)
+	}
+	defer func() {
+		if err := os.Remove(apiserverConfig.Name()); err != nil {
+			t.Errorf("Failed to remove config file: %v", err)
+		}
+	}()
+
+	configStr := fmt.Sprintf(`
+apiVersion: v1
+kind: Config
+clusters:
+- cluster:
+    server: %s
+    certificate-authority: %s
+  name: integration
+contexts:
+- context:
+    cluster: integration
+    user: kube-scheduler
+  name: default-context
+current-context: default-context
+users:
+- name: kube-scheduler
+  user:
+    token: fake-token
+`, s.ClientConfig.Host, s.ServerOpts.SecureServing.ServerCert.CertKey.CertFile)
+
+	if _, err = apiserverConfig.WriteString(configStr); err != nil {
+		t.Fatalf("Failed to write config file: %v", err)
+	}
+
+	_, ctx := ktesting.NewTestContext(t)
+	schedulerServer, err := kubeschedulertesting.StartTestServer(
+		t, ctx,
+		[]string{"--kubeconfig", apiserverConfig.Name(), "--leader-elect=false", "--authentication-skip-lookup=true", "--authorization-always-allow-paths=/metrics"},
+	)
+	if err != nil {
+		t.Fatalf("Failed to start kube-scheduler server: %v", err)
+	}
+	if schedulerServer.TearDownFn != nil {
+		defer schedulerServer.TearDownFn()
+	}
+
+	secureInfo := schedulerServer.Config.SecureServing
+	secureOptions := schedulerServer.Options.SecureServing
+	url := fmt.Sprintf("https://%s", secureInfo.Listener.Addr().String())
+	url = strings.ReplaceAll(url, "[::]", "127.0.0.1")
+
+	pool := x509.NewCertPool()
+	serverCertPath := path.Join(secureOptions.ServerCert.CertDirectory, secureOptions.ServerCert.PairName+".crt")
+	serverCert, err := os.ReadFile(serverCertPath)
+	if err != nil {
+		t.Fatalf("Failed to read component server cert: %v", err)
+	}
+	pool.AppendCertsFromPEM(serverCert)
+	httpClient := &http.Client{
+		Transport: &http.Transport{
+			TLSClientConfig: &tls.Config{RootCAs: pool},
+		},
+	}
+
+	histogramMetric := "scheduler_scheduling_algorithm_duration_seconds"
+	metrics, err := testutil.ScrapeMetricsProto(url+"/metrics", httpClient)
+	if err != nil {
+		t.Fatalf("failed to scrape metrics: %v", err)
+	}
+
+	mf, ok := metrics[histogramMetric]
+	if !ok {
+		t.Fatalf("metric %q not found", histogramMetric)
+	}
+
+	testutil.AssertHasNativeHistogram(t, mf, nil)
+}