deflake restart count assertions in in-place resize tests (kubernetes#131055)

Author: natasha41575

test/e2e/common/node/pod_resize.go

@@ -1200,13 +1200,14 @@ func doPodResizeTests() {
 				patchedPod, pErr = f.ClientSet.CoreV1().Pods(newPod.Namespace).Patch(ctx, newPod.Name,
 					types.StrategicMergePatchType, []byte(patchString), metav1.PatchOptions{}, "resize")
 				framework.ExpectNoError(pErr, fmt.Sprintf("failed to patch pod for %s", opStr))
+				expected := e2epod.UpdateExpectedContainerRestarts(ctx, patchedPod, expectedContainers)
 
 				ginkgo.By(fmt.Sprintf("verifying pod patched for %s", opStr))
-				e2epod.VerifyPodResources(patchedPod, expectedContainers)
+				e2epod.VerifyPodResources(patchedPod, expected)
 
 				ginkgo.By(fmt.Sprintf("waiting for %s to be actuated", opStr))
-				resizedPod := e2epod.WaitForPodResizeActuation(ctx, f, podClient, newPod, expectedContainers)
-				e2epod.ExpectPodResized(ctx, f, resizedPod, expectedContainers)
+				resizedPod := e2epod.WaitForPodResizeActuation(ctx, f, podClient, newPod, expected)
+				e2epod.ExpectPodResized(ctx, f, resizedPod, expected)
 			}
 
 			patchAndVerify(tc.patchString, tc.expected, "resize")
@@ -1219,7 +1220,7 @@ func doPodResizeTests() {
 					gomega.Expect(c.Name).To(gomega.Equal(tc.expected[i].Name),
 						"test case containers & expectations should be in the same order")
 					// Resizes that trigger a restart should trigger a second restart when rolling back.
-					rollbackContainers[i].RestartCount = tc.expected[i].RestartCount * 2
+					rollbackContainers[i].RestartCount = tc.expected[i].RestartCount
 				}
 
 				rbPatchStr, err := e2epod.ResizeContainerPatch(tc.containers)

test/e2e/framework/pod/resize.go

@@ -505,6 +505,24 @@ func ResizeContainerPatch(containers []ResizableContainerInfo) (string, error) {
 	return string(patchBytes), nil
 }
 
+// UpdateExpectedContainerRestarts updates the RestartCounts in expectedContainers by
+// adding them to the existing RestartCounts in the containerStatuses of the provided pod.
+// This reduces the flakiness of the RestartCount assertions by grabbing the current
+// restart count right before the resize operation, and verify the expected increment (0 or 1)
+// rather than the absolute count.
+func UpdateExpectedContainerRestarts(ctx context.Context, pod *v1.Pod, expectedContainers []ResizableContainerInfo) []ResizableContainerInfo {
+	initialRestarts := make(map[string]int32)
+	newExpectedContainers := []ResizableContainerInfo{}
+	for _, ctr := range pod.Status.ContainerStatuses {
+		initialRestarts[ctr.Name] = ctr.RestartCount
+	}
+	for i, ctr := range expectedContainers {
+		newExpectedContainers = append(newExpectedContainers, expectedContainers[i])
+		newExpectedContainers[i].RestartCount += initialRestarts[ctr.Name]
+	}
+	return newExpectedContainers
+}
+
 func formatErrors(err error) error {
 	// Put each error on a new line for readability.
 	var agg utilerrors.Aggregate

test/e2e/node/pod_resize.go

@@ -155,6 +155,7 @@ func doPodResizeAdmissionPluginsTests() {
 			patchedPod, pErr := f.ClientSet.CoreV1().Pods(newPods[0].Namespace).Patch(ctx, newPods[0].Name,
 				types.StrategicMergePatchType, []byte(patchString), metav1.PatchOptions{}, "resize")
 			framework.ExpectNoError(pErr, "failed to patch pod for resize")
+			expected = e2epod.UpdateExpectedContainerRestarts(ctx, patchedPod, expected)
 
 			ginkgo.By("verifying pod patched for resize within resource quota")
 			e2epod.VerifyPodResources(patchedPod, expected)
@@ -405,8 +406,9 @@ func doPodResizeSchedulerTests(f *framework.Framework) {
 		ginkgo.By(fmt.Sprintf("TEST3: Verify pod '%s' is resized successfully after pod deletion '%s' and '%s", testPod1.Name, testPod2.Name, testPod3.Name))
 		expected := []e2epod.ResizableContainerInfo{
 			{
-				Name:      "c1",
-				Resources: &e2epod.ContainerResources{CPUReq: testPod1CPUQuantity.String(), CPULim: testPod1CPUQuantity.String()},
+				Name:         "c1",
+				Resources:    &e2epod.ContainerResources{CPUReq: testPod1CPUQuantity.String(), CPULim: testPod1CPUQuantity.String()},
+				RestartCount: testPod1.Status.ContainerStatuses[0].RestartCount,
 			},
 		}
 		resizedPod := e2epod.WaitForPodResizeActuation(ctx, f, podClient, testPod1, expected)