[Perf] Optimize and fix informer leak in core/bench benchmark tests

cluster-autoscaler/cloudprovider/test/fake_cloud_provider.go

@@ -346,6 +346,9 @@ func (n *NodeGroup) IncreaseSize(delta int) error {
 	if n.targetSize+delta > n.maxSize {
 		return fmt.Errorf("size too large")
 	}
+	if n.template == nil || n.template.Node() == nil {
+		return fmt.Errorf("node group %s has no template to create new nodes", n.id)
+	}
 
 	n.provider.Lock()
 	defer n.provider.Unlock()
@@ -354,9 +357,6 @@ func (n *NodeGroup) IncreaseSize(delta int) error {
 		instanceNum := n.targetSize + i
 		instanceId := fmt.Sprintf("%s-node-%d", n.id, instanceNum)
 
-		if n.template == nil || n.template.Node() == nil {
-			return fmt.Errorf("node group %s has no template to create new nodes", n.id)
-		}
 		newNode := n.template.Node().DeepCopy()
 		newNode.Name = instanceId
 
@@ -370,6 +370,39 @@ func (n *NodeGroup) IncreaseSize(delta int) error {
 	return nil
 }
 
+// IncreaseSizeDirect is a performance-optimized version of IncreaseSize that adds nodes
+// directly to the fake client's object tracker to bypass API serialization and watch event overhead.
+// Only safe to call before Informer Factory starts.
+func (n *NodeGroup) IncreaseSizeDirect(delta int) error {
+	n.Lock()
+	defer n.Unlock()
+	if n.targetSize+delta > n.maxSize {
+		return fmt.Errorf("size too large")
+	}
+	if n.template == nil || n.template.Node() == nil {
+		return fmt.Errorf("node group %s has no template to create new nodes", n.id)
+	}
+
+	n.provider.Lock()
+	defer n.provider.Unlock()
+
+	for i := 0; i < delta; i++ {
+		instanceNum := n.targetSize + i
+		instanceId := fmt.Sprintf("%s-node-%d", n.id, instanceNum)
+
+		newNode := n.template.Node().DeepCopy()
+		newNode.Name = instanceId
+
+		n.instances[instanceId] = cloudprovider.InstanceRunning
+		n.provider.nodeToGroup[instanceId] = n.id
+		if n.provider.k8s != nil {
+			n.provider.k8s.AddNodeDirect(newNode)
+		}
+	}
+	n.targetSize += delta
+	return nil
+}
+
 // TemplateNodeInfo returns the template node information for this node group.
 func (n *NodeGroup) TemplateNodeInfo() (*framework.NodeInfo, error) {
 	if n.template == nil {

cluster-autoscaler/core/bench/benchmark_runonce_test.go

@@ -133,18 +133,21 @@
 	}
 
 	for i := 0; i < b.N; i++ {
+		ctx, cancel := context.WithCancel(context.Background())
 		clusterFakes := newClusterFakes()
 		if err := s.setup(clusterFakes); err != nil {
+			cancel()
 			b.Fatalf("setup failed: %v", err)
 		}
-		autoscaler := newAutoscaler(b, s, clusterFakes)
+		autoscaler := newAutoscaler(b, s, clusterFakes, ctx)
 
 		// Manually trigger GC before the timed section to ensure a clean state
 		// for each iteration.
 		runtime.GC()
 
 		if f != nil && i == 0 {
 			if err := pprof.StartCPUProfile(f); err != nil {
+				cancel()
 				b.Fatalf("Failed to start cpu profile: %v", err)
 			}
 		}
@@ -158,14 +161,17 @@
 		}
 
 		if err != nil {
+			cancel()
 			b.Fatalf("RunOnce failed: %v", err)
 		}
 
 		if s.verify != nil {
 			if err := s.verify(clusterFakes); err != nil {
+				cancel()
 				b.Fatalf("verify failed: %v", err)
 			}
 		}
+		cancel()
 	}
 }
 
@@ -178,7 +184,7 @@
 }
 
 // newAutoscaler constructs a core.Autoscaler instance configured for the given scenario.
-func newAutoscaler(b *testing.B, s scenario, clusterFakes *integration.FakeSet) core.Autoscaler {
+func newAutoscaler(b *testing.B, s scenario, clusterFakes *integration.FakeSet, ctx context.Context) core.Autoscaler {
 	opts := defaultCAOptions()
 	if s.config != nil {
 		s.config(&opts)
@@ -190,7 +196,7 @@
 	ftkc := &fastTaintingKubeClient{taintedNodes: make(map[string]bool)}
 	ftkc.registerReactors(clusterFakes.KubeClient)
 
-	kubeClients := ca_context.NewAutoscalingKubeClients(context.Background(), opts, clusterFakes.KubeClient, clusterFakes.InformerFactory)
+	kubeClients := ca_context.NewAutoscalingKubeClients(ctx, opts, clusterFakes.KubeClient, clusterFakes.InformerFactory)
 	kubeClients.Recorder = &noOpRecorder{}
 
 	wrappedCloudProvider := &fastScaleUpCloudProvider{
@@ -204,7 +210,7 @@
 		WithAutoscalingKubeClients(kubeClients).
 		WithInformerFactory(clusterFakes.InformerFactory).
 		WithCloudProvider(wrappedCloudProvider).
-		WithPodObserver(clusterFakes.PodObserver).Build(context.Background())
+		WithPodObserver(clusterFakes.PodObserver).Build(ctx)
 	if err != nil {
 		b.Fatalf("Failed to build: %v", err)
 	}
@@ -381,7 +387,7 @@
 			cpu := int64(nodeCPU / podsPerNode)
 			mem := int64(nodeMem / podsPerNode)
 			pod := BuildTestPod(podName, cpu, mem, MarkUnschedulable())
-			clusterFakes.K8s.AddPod(pod)
+			clusterFakes.K8s.AddPodDirect(pod)
 		}
 		return nil
 	}
@@ -401,8 +407,8 @@
 			testprovider.WithNGSize(0, maxNGSize),
 		)
 
-		ng := clusterFakes.CloudProvider.GetNodeGroup(ngName)
-		if err := ng.IncreaseSize(nodesCount); err != nil {
+		ng := clusterFakes.CloudProvider.GetNodeGroup(ngName).(*testprovider.NodeGroup)
+		if err := ng.IncreaseSizeDirect(nodesCount); err != nil {
 			return err
 		}
 
@@ -420,7 +426,7 @@
 				pod.Annotations = make(map[string]string)
 			}
 			pod.Annotations["cluster-autoscaler.kubernetes.io/safe-to-evict"] = "true"
-			clusterFakes.K8s.AddPod(pod)
+			clusterFakes.K8s.AddPodDirect(pod)
 		}
 
 		return nil

cluster-autoscaler/utils/fake/kubernetes.go

@@ -45,6 +45,19 @@ func (k *Kubernetes) AddNode(node *apiv1.Node) {
 	_, _ = k.Client.CoreV1().Nodes().Create(context.TODO(), node, metav1.CreateOptions{})
 }
 
+// AddNodeDirect adds a node directly to the fake client's underlying object tracker
+// via Tracker().Add(node) instead of going through CoreV1().Nodes().Create().
+// This completely bypasses the fake clientset's serialization, deep-copy, and reactor overhead,
+// reducing setup times for large-scale simulations (like benchmarks) by orders of magnitude.
+//
+// WARNING: Because Tracker().Add() does NOT trigger active watch events in client-go,
+// this method MUST ONLY be used for pre-populating the cluster state during the initial
+// setup phase BEFORE the Informer Factory is started.
+// To dynamically add a node AFTER informers are running, use the standard AddNode method instead.
+func (k *Kubernetes) AddNodeDirect(node *apiv1.Node) {
+	_ = k.Client.Tracker().Add(node)
+}
+
 // UpdateNode updates a node.
 func (k *Kubernetes) UpdateNode(node *apiv1.Node) {
 	_, _ = k.Client.CoreV1().Nodes().Update(context.TODO(), node, metav1.UpdateOptions{})
@@ -66,6 +79,19 @@ func (k *Kubernetes) AddPod(pod *apiv1.Pod) {
 	_, _ = k.Client.CoreV1().Pods(pod.Namespace).Create(context.TODO(), pod, metav1.CreateOptions{})
 }
 
+// AddPodDirect adds a pod directly to the fake client's underlying object tracker
+// via Tracker().Add(pod) instead of going through CoreV1().Pods().Create().
+// This completely bypasses the fake clientset's serialization, deep-copy, and reactor overhead,
+// reducing setup times for large-scale simulations (like benchmarks) by orders of magnitude.
+//
+// WARNING: Because Tracker().Add() does NOT trigger active watch events in client-go,
+// this method MUST ONLY be used for pre-populating the cluster state during the initial
+// setup phase BEFORE the Informer Factory is started.
+// To dynamically add a pod AFTER informers are running, use the standard AddPod method instead.
+func (k *Kubernetes) AddPodDirect(pod *apiv1.Pod) {
+	_ = k.Client.Tracker().Add(pod)
+}
+
 // DeletePod deletes a pod.
 func (k *Kubernetes) DeletePod(namespace, name string) {
 	_ = k.Client.CoreV1().Pods(namespace).Delete(context.TODO(), name, metav1.DeleteOptions{})