Merge pull request #8052 from plkokanov/fix/nil-pointer

VPA: Fix nil ptr when loading metrics and `memory-saver=true`

Author: k8s-ci-robot

vertical-pod-autoscaler/pkg/recommender/input/cluster_feeder.go

@@ -499,15 +499,17 @@ func (feeder *clusterStateFeeder) LoadRealTimeMetrics(ctx context.Context) {
 	sampleCount := 0
 	droppedSampleCount := 0
 	for _, containerMetrics := range containersMetrics {
-		podInitContainers := feeder.clusterState.Pods()[containerMetrics.ID.PodID].InitContainers
-		if slices.Contains(podInitContainers, containerMetrics.ID.ContainerName) {
-			klog.V(3).InfoS("Skipping metric samples for init container", "pod", klog.KRef(containerMetrics.ID.PodID.Namespace, containerMetrics.ID.PodID.PodName), "container", containerMetrics.ID.ContainerName)
-			droppedSampleCount += len(containerMetrics.Usage)
-			continue
+		// Container metrics are fetched for all pods, however, not all pod states are tracked in memory saver mode.
+		if pod, exists := feeder.clusterState.Pods()[containerMetrics.ID.PodID]; exists && pod != nil {
+			if slices.Contains(pod.InitContainers, containerMetrics.ID.ContainerName) {
+				klog.V(3).InfoS("Skipping metric samples for init container", "pod", klog.KRef(containerMetrics.ID.PodID.Namespace, containerMetrics.ID.PodID.PodName), "container", containerMetrics.ID.ContainerName)
+				droppedSampleCount += len(containerMetrics.Usage)
+				continue
+			}
 		}
 		for _, sample := range newContainerUsageSamplesWithKey(containerMetrics) {
 			if err := feeder.clusterState.AddSample(sample); err != nil {
-				// Not all pod states are tracked in memory saver mode
+				// Not all pod states are tracked in memory saver mode.
 				if _, isKeyError := err.(model.KeyError); isKeyError && feeder.memorySaveMode {
 					continue
 				}

vertical-pod-autoscaler/pkg/recommender/input/cluster_feeder_test.go

@@ -97,6 +97,10 @@ type fakeClusterState struct {
 }
 
 func (cs *fakeClusterState) AddSample(sample *model.ContainerUsageSampleWithKey) error {
+	_, podExists := cs.stubbedPods[sample.Container.PodID]
+	if !podExists {
+		return model.NewKeyError(sample.Container.PodID)
+	}
 	samplesForContainer := cs.addedSamples[sample.Container]
 	cs.addedSamples[sample.Container] = append(samplesForContainer, sample)
 	return nil
@@ -651,6 +655,27 @@ func TestClusterStateFeeder_LoadRealTimeMetrics(t *testing.T) {
 	samplesForContainer2 := clusterState.addedSamples[regularContainer2]
 	assert.Contains(t, samplesForContainer2, regularContainer2UsageSamples[0])
 	assert.Contains(t, samplesForContainer2, regularContainer2UsageSamples[1])
+
+	// Add extra container metrics for which there are no added pods to the state to simulate memory-saver=true
+	extraPodID := model.PodID{Namespace: namespaceName, PodName: "ExtraPod"}
+	extraContainer := model.ContainerID{PodID: extraPodID, ContainerName: "ExtraContainer"}
+	extraContainerMetricsSnapshot, _ := newContainerMetricsSnapshot(extraContainer, 200, 2048)
+	containerMetricsSnapshots = append(containerMetricsSnapshots, extraContainerMetricsSnapshot)
+
+	clusterState = NewFakeClusterState(nil, pods)
+
+	feeder = clusterStateFeeder{
+		memorySaveMode: true,
+		clusterState:   clusterState,
+		metricsClient:  fakeMetricsClient{snapshots: containerMetricsSnapshots},
+	}
+
+	feeder.LoadRealTimeMetrics(tctx)
+
+	assert.Equal(t, 2, len(clusterState.addedSamples))
+
+	_, samplesForExtraContainerExist := clusterState.addedSamples[extraContainer]
+	assert.False(t, samplesForExtraContainerExist)
 }
 
 type fakeHistoryProvider struct {