Merge pull request #113669 from wojtek-t/clean_shutdown_final

Clean shutdown final

test/integration/apiserver/apply/apply_crd_test.go

@@ -435,6 +435,7 @@ func TestApplyCRDUnhandledSchema(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
+	defer etcdclient.Close()
 
 	server, err := apiservertesting.StartTestServer(t, apiservertesting.NewDefaultTestServerOptions(), nil, storageConfig)
 	if err != nil {

test/integration/framework/etcd.go

@@ -194,11 +194,13 @@ func EtcdMain(tests func() int) {
 	stop() // Don't defer this. See os.Exit documentation.
 
 	checkNumberOfGoroutines := func() (bool, error) {
-		// Leave some room for goroutines we can not get rid of
-		// like k8s.io/klog/v2.(*loggingT).flushDaemon()
-		// TODO(#108483): Figure out if we can reduce this
-		//   further (ideally down to zero).
-		if dg := runtime.NumGoroutine() - before; dg <= 4 {
+		// We leave some room for leaked goroutines as there are
+		// still some leaks, mostly:
+		// - leak from lumberjack package we're vendoring
+		// - leak from apiserve healthz
+		// - leak from opencensus library
+		// Once fixed, we should be able to bring it down to zero.
+		if dg := runtime.NumGoroutine() - before; dg <= 3 {
 			return true, nil
 		}
 		// Allow goroutines to schedule and die off.
@@ -210,7 +212,9 @@ func EtcdMain(tests func() int) {
 	// But we keep the limit higher to account for cpu-starved environments.
 	if err := wait.Poll(100*time.Millisecond, 5*time.Second, checkNumberOfGoroutines); err != nil {
 		after := runtime.NumGoroutine()
-		klog.Fatalf("unexpected number of goroutines: before: %d after %d", before, after)
+		stacktraces := make([]byte, 1<<20)
+		runtime.Stack(stacktraces, true)
+		klog.Fatalf("unexpected number of goroutines: before: %d after %d\n%sd", before, after, string(stacktraces))
 	}
 	os.Exit(result)
 }

test/integration/namespace/ns_conditions_test.go

@@ -119,8 +119,15 @@ func TestNamespaceCondition(t *testing.T) {
 
 // TestNamespaceLabels tests for default labels added in https://github.com/kubernetes/kubernetes/pull/96968
 func TestNamespaceLabels(t *testing.T) {
-	closeFn, _, _, kubeClient, _ := namespaceLifecycleSetup(t)
+	closeFn, nsController, _, kubeClient, _ := namespaceLifecycleSetup(t)
 	defer closeFn()
+
+	// Even though nscontroller isn't used in this test, its creation is already
+	// spawning some goroutines. So we need to run it to ensure they won't leak.
+	stopCh := make(chan struct{})
+	close(stopCh)
+	go nsController.Run(5, stopCh)
+
 	nsName := "test-namespace-labels-generated"
 	// Create a new namespace w/ no name
 	ns, err := kubeClient.CoreV1().Namespaces().Create(context.TODO(), &corev1.Namespace{

test/integration/scheduler/queue_test.go

@@ -63,6 +63,7 @@ func TestCoreResourceEnqueue(t *testing.T) {
 	testutils.SyncInformerFactory(testCtx)
 
 	defer testutils.CleanupTest(t, testCtx)
+	defer testCtx.Scheduler.SchedulingQueue.Close()
 
 	cs, ns, ctx := testCtx.ClientSet, testCtx.NS.Name, testCtx.Ctx
 	// Create one Node with a taint.

test/integration/util/util.go

@@ -899,7 +899,7 @@ func timeout(ctx context.Context, d time.Duration, f func()) error {
 	done := make(chan struct{})
 	go func() {
 		f()
-		done <- struct{}{}
+		close(done)
 	}()
 
 	select {