Merge pull request #138087 from dims/fix-memory-manager-metrics-cleanup

e2e_node: wait for pod drain before asserting zero pods in Memory Manager Metrics
This commit is contained in:
Kubernetes Prow Robot
2026-03-30 18:00:15 +05:30
committed by GitHub

View File

@@ -79,8 +79,21 @@ var _ = SIGDescribe("Memory Manager Metrics", framework.WithSerial(), feature.Me
updateKubeletConfig(ctx, f, oldCfg, true)
})
count := printAllPodsOnNode(ctx, f.ClientSet, framework.TestContext.NodeName)
gomega.Expect(count).To(gomega.BeZero(), "unexpected pods on %q, please check output above", framework.TestContext.NodeName)
// Wait for pods from previous serial tests to finish terminating.
// Framework namespace deletion is async, and the kubelet restart
// above can delay in-flight pod cleanup. Poll instead of asserting
// immediately so we don't flake on test ordering.
gomega.Eventually(ctx, func(ctx context.Context) error {
count, err := printAllPodsOnNode(ctx, f.ClientSet, framework.TestContext.NodeName)
if err != nil {
return err
}
if count != 0 {
return fmt.Errorf("unexpected pods on %q: %d still present", framework.TestContext.NodeName, count)
}
return nil
}).WithTimeout(2*time.Minute).WithPolling(5*time.Second).Should(
gomega.Succeed(), "unexpected pods on %q, please check output above", framework.TestContext.NodeName)
})
ginkgo.It("should report zero pinning counters after a fresh restart", func(ctx context.Context) {
@@ -160,7 +173,8 @@ var _ = SIGDescribe("Memory Manager Metrics", framework.WithSerial(), feature.Me
}),
)
printAllPodsOnNode(ctx, f.ClientSet, framework.TestContext.NodeName)
_, err := printAllPodsOnNode(ctx, f.ClientSet, framework.TestContext.NodeName)
framework.ExpectNoError(err)
// we updated the kubelet config in BeforeEach, so we can assume we start fresh.
// being [Serial], we can also assume noone else but us is running pods.
@@ -213,7 +227,7 @@ func validateMetrics(values e2emetrics.KubeletMetrics, totalKey, errorKey string
// printAllPodsOnNode outputs status of all kubelet pods into log.
// Note considering the e2e_node environment we will always have exactly 1 node, but still.
func printAllPodsOnNode(ctx context.Context, c clientset.Interface, nodeName string) int {
func printAllPodsOnNode(ctx context.Context, c clientset.Interface, nodeName string) (int, error) {
nodeSelector := fields.Set{
"spec.nodeName": nodeName,
}.AsSelector().String()
@@ -223,7 +237,7 @@ func printAllPodsOnNode(ctx context.Context, c clientset.Interface, nodeName str
})
if err != nil {
framework.Logf("Unable to retrieve pods for node %v: %v", nodeName, err)
return 0
return 0, err
}
count := 0
framework.Logf("begin listing pods: %d found", len(podList.Items))
@@ -237,5 +251,5 @@ func printAllPodsOnNode(ctx context.Context, c clientset.Interface, nodeName str
count++
}
framework.Logf("end listing pods: %d found", len(podList.Items))
return count
return count, nil
}