Merge pull request #90448 from soltysh/issue87851

Fix run command when waiting for pods

staging/src/k8s.io/kubectl/pkg/cmd/debug/debug.go

@@ -355,21 +355,6 @@ func waitForEphemeralContainer(ctx context.Context, podClient corev1client.PodsG
 	ctx, cancel := watchtools.ContextWithOptionalTimeout(ctx, 0*time.Second)
 	defer cancel()
 
-	preconditionFunc := func(store cache.Store) (bool, error) {
-		_, exists, err := store.Get(&metav1.ObjectMeta{Namespace: ns, Name: podName})
-		if err != nil {
-			return true, err
-		}
-		if !exists {
-			// We need to make sure we see the object in the cache before we start waiting for events
-			// or we would be waiting for the timeout if such object didn't exist.
-			// (e.g. it was deleted before we started informers so they wouldn't even see the delete event)
-			return true, errors.NewNotFound(corev1.Resource("pods"), podName)
-		}
-
-		return false, nil
-	}
-
 	fieldSelector := fields.OneTermEqualSelector("metadata.name", podName).String()
 	lw := &cache.ListWatch{
 		ListFunc: func(options metav1.ListOptions) (runtime.Object, error) {
@@ -385,7 +370,7 @@ func waitForEphemeralContainer(ctx context.Context, podClient corev1client.PodsG
 	intr := interrupt.New(nil, cancel)
 	var result *corev1.Pod
 	err := intr.Run(func() error {
-		ev, err := watchtools.UntilWithSync(ctx, lw, &corev1.Pod{}, preconditionFunc, func(ev watch.Event) (bool, error) {
+		ev, err := watchtools.UntilWithSync(ctx, lw, &corev1.Pod{}, nil, func(ev watch.Event) (bool, error) {
 			switch ev.Type {
 			case watch.Deleted:
 				return false, errors.NewNotFound(schema.GroupResource{Resource: "pods"}, "")

staging/src/k8s.io/kubectl/pkg/cmd/rollout/BUILD

@@ -21,7 +21,6 @@ go_library(
         "//build/visible_to:pkg_kubectl_cmd_rollout_CONSUMERS",
     ],
     deps = [
-        "//staging/src/k8s.io/apimachinery/pkg/api/errors:go_default_library",
         "//staging/src/k8s.io/apimachinery/pkg/api/meta:go_default_library",
         "//staging/src/k8s.io/apimachinery/pkg/apis/meta/v1:go_default_library",
         "//staging/src/k8s.io/apimachinery/pkg/apis/meta/v1/unstructured:go_default_library",

staging/src/k8s.io/kubectl/pkg/cmd/rollout/rollout_status.go

@@ -23,7 +23,6 @@ import (
 
 	"github.com/spf13/cobra"
 
-	apierrors "k8s.io/apimachinery/pkg/api/errors"
 	"k8s.io/apimachinery/pkg/api/meta"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
@@ -200,25 +199,11 @@ func (o *RolloutStatusOptions) Run() error {
 		},
 	}
 
-	preconditionFunc := func(store cache.Store) (bool, error) {
-		_, exists, err := store.Get(&metav1.ObjectMeta{Namespace: info.Namespace, Name: info.Name})
-		if err != nil {
-			return true, err
-		}
-		if !exists {
-			// We need to make sure we see the object in the cache before we start waiting for events
-			// or we would be waiting for the timeout if such object didn't exist.
-			return true, apierrors.NewNotFound(mapping.Resource.GroupResource(), info.Name)
-		}
-
-		return false, nil
-	}
-
 	// if the rollout isn't done yet, keep watching deployment status
 	ctx, cancel := watchtools.ContextWithOptionalTimeout(context.Background(), o.Timeout)
 	intr := interrupt.New(nil, cancel)
 	return intr.Run(func() error {
-		_, err = watchtools.UntilWithSync(ctx, lw, &unstructured.Unstructured{}, preconditionFunc, func(e watch.Event) (bool, error) {
+		_, err = watchtools.UntilWithSync(ctx, lw, &unstructured.Unstructured{}, nil, func(e watch.Event) (bool, error) {
 			switch t := e.Type; t {
 			case watch.Added, watch.Modified:
 				status, done, err := statusViewer.Status(e.Object.(runtime.Unstructured), o.Revision)

staging/src/k8s.io/kubectl/pkg/cmd/run/run.go

@@ -381,17 +381,21 @@ func (o *RunOptions) Run(f cmdutil.Factory, cmd *cobra.Command, args []string) e
 		}
 
 		var pod *corev1.Pod
-		leaveStdinOpen := o.LeaveStdinOpen
+		waitForExitCode := !o.LeaveStdinOpen && (restartPolicy == corev1.RestartPolicyNever || restartPolicy == corev1.RestartPolicyOnFailure)
-		waitForExitCode := !leaveStdinOpen && restartPolicy == corev1.RestartPolicyNever
 		if waitForExitCode {
-			pod, err = waitForPod(clientset.CoreV1(), attachablePod.Namespace, attachablePod.Name, opts.GetPodTimeout, podCompleted)
+			// we need different exit condition depending on restart policy
+			// for Never it can either fail or succeed, for OnFailure only
+			// success matters
+			exitCondition := podCompleted
+			if restartPolicy == corev1.RestartPolicyOnFailure {
+				exitCondition = podSucceeded
+			}
+			pod, err = waitForPod(clientset.CoreV1(), attachablePod.Namespace, attachablePod.Name, opts.GetPodTimeout, exitCondition)
 			if err != nil {
 				return err
 			}
-		}
+		} else {
-
 			// after removal is done, return successfully if we are not interested in the exit code
-		if !waitForExitCode {
 			return nil
 		}
 
@@ -457,21 +461,6 @@ func waitForPod(podClient corev1client.PodsGetter, ns, name string, timeout time
 	ctx, cancel := watchtools.ContextWithOptionalTimeout(context.Background(), timeout)
 	defer cancel()
 
-	preconditionFunc := func(store cache.Store) (bool, error) {
-		_, exists, err := store.Get(&metav1.ObjectMeta{Namespace: ns, Name: name})
-		if err != nil {
-			return true, err
-		}
-		if !exists {
-			// We need to make sure we see the object in the cache before we start waiting for events
-			// or we would be waiting for the timeout if such object didn't exist.
-			// (e.g. it was deleted before we started informers so they wouldn't even see the delete event)
-			return true, errors.NewNotFound(corev1.Resource("pods"), name)
-		}
-
-		return false, nil
-	}
-
 	fieldSelector := fields.OneTermEqualSelector("metadata.name", name).String()
 	lw := &cache.ListWatch{
 		ListFunc: func(options metav1.ListOptions) (runtime.Object, error) {
@@ -487,9 +476,7 @@ func waitForPod(podClient corev1client.PodsGetter, ns, name string, timeout time
 	intr := interrupt.New(nil, cancel)
 	var result *corev1.Pod
 	err := intr.Run(func() error {
-		ev, err := watchtools.UntilWithSync(ctx, lw, &corev1.Pod{}, preconditionFunc, func(ev watch.Event) (bool, error) {
+		ev, err := watchtools.UntilWithSync(ctx, lw, &corev1.Pod{}, nil, exitCondition)
-			return exitCondition(ev)
-		})
 		if ev != nil {
 			result = ev.Object.(*corev1.Pod)
 		}
@@ -708,6 +695,20 @@ func podCompleted(event watch.Event) (bool, error) {
 	return false, nil
 }
 
+// podSucceeded returns true if the pod has run to completion, false if the pod has not yet
+// reached running state, or an error in any other case.
+func podSucceeded(event watch.Event) (bool, error) {
+	switch event.Type {
+	case watch.Deleted:
+		return false, errors.NewNotFound(schema.GroupResource{Resource: "pods"}, "")
+	}
+	switch t := event.Object.(type) {
+	case *corev1.Pod:
+		return t.Status.Phase == corev1.PodSucceeded, nil
+	}
+	return false, nil
+}
+
 // podRunningAndReady returns true if the pod is running and ready, false if the pod has not
 // yet reached those states, returns ErrPodCompleted if the pod has run to completion, or
 // an error in any other case.

test/e2e/kubectl/kubectl.go

@@ -522,13 +522,21 @@ var _ = SIGDescribe("Kubectl client", func() {
 			_, err = framework.NewKubectlCommand(ns, nsFlag, "run", "-i", "--image="+busyboxImage, "--restart=OnFailure", "failure-2", "--", "/bin/sh", "-c", "cat && exit 42").
 				WithStdinData("abcd1234").
 				Exec()
-			framework.ExpectNoError(err)
+			ee, ok = err.(uexec.ExitError)
+			framework.ExpectEqual(ok, true)
+			if !strings.Contains(ee.String(), "timed out") {
+				framework.Failf("Missing expected 'timed out' error, got: %#v", ee)
+			}
 
 			ginkgo.By("running a failing command without --restart=Never, but with --rm")
 			_, err = framework.NewKubectlCommand(ns, nsFlag, "run", "-i", "--image="+busyboxImage, "--restart=OnFailure", "--rm", "failure-3", "--", "/bin/sh", "-c", "cat && exit 42").
 				WithStdinData("abcd1234").
 				Exec()
-			framework.ExpectNoError(err)
+			ee, ok = err.(uexec.ExitError)
+			framework.ExpectEqual(ok, true)
+			if !strings.Contains(ee.String(), "timed out") {
+				framework.Failf("Missing expected 'timed out' error, got: %#v", ee)
+			}
 			e2epod.WaitForPodToDisappear(f.ClientSet, ns, "failure-3", labels.Everything(), 2*time.Second, wait.ForeverTestTimeout)
 
 			ginkgo.By("running a failing command with --leave-stdin-open")