diff --git a/pkg/kubelet/volumemanager/populator/desired_state_of_world_populator.go b/pkg/kubelet/volumemanager/populator/desired_state_of_world_populator.go index 63c7152040e..75d1a1e833b 100644 --- a/pkg/kubelet/volumemanager/populator/desired_state_of_world_populator.go +++ b/pkg/kubelet/volumemanager/populator/desired_state_of_world_populator.go @@ -24,6 +24,7 @@ import ( "context" "errors" "fmt" + "slices" "sync" "time" @@ -528,15 +529,21 @@ func (dswp *desiredStateOfWorldPopulator) getPVCExtractPV( return nil, fmt.Errorf("failed to fetch PVC from API server: %v", err) } - // Pods that uses a PVC that is being deleted must not be started. + // Pods that uses a PVC that is being deleted and not protected by + // kubernetes.io/pvc-protection must not be started. // - // In case an old kubelet is running without this check or some kubelets - // have this feature disabled, the worst that can happen is that such - // pod is scheduled. This was the default behavior in 1.8 and earlier - // and users should not be that surprised. + // 1) In case an old kubelet is running without this check, the worst + // that can happen is that such pod is scheduled. This was the default + // behavior in 1.8 and earlier and users should not be that surprised. // It should happen only in very rare case when scheduler schedules // a pod and user deletes a PVC that's used by it at the same time. - if pvc.ObjectMeta.DeletionTimestamp != nil { + // + // 2) Adding a check for kubernetes.io/pvc-protection here to prevent + // the existing running pods from being affected during the rebuild of + // the desired state of the world cache when the kubelet is restarted. + // It is safe for kubelet to add this check here because the PVC will + // be stuck in Terminating state until the pod is deleted. + if pvc.ObjectMeta.DeletionTimestamp != nil && !slices.Contains(pvc.Finalizers, util.PVCProtectionFinalizer) { return nil, errors.New("PVC is being deleted") } diff --git a/test/e2e/storage/csimock/csi_kubelet_restart.go b/test/e2e/storage/csimock/csi_kubelet_restart.go new file mode 100644 index 00000000000..105ac427640 --- /dev/null +++ b/test/e2e/storage/csimock/csi_kubelet_restart.go @@ -0,0 +1,100 @@ +/* +Copyright 2025 The Kubernetes Authors. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +package csimock + +import ( + "context" + "fmt" + + "github.com/onsi/ginkgo/v2" + "github.com/onsi/gomega" + apierrors "k8s.io/apimachinery/pkg/api/errors" + metav1 "k8s.io/apimachinery/pkg/apis/meta/v1" + "k8s.io/kubernetes/test/e2e/framework" + e2epod "k8s.io/kubernetes/test/e2e/framework/pod" + e2eskipper "k8s.io/kubernetes/test/e2e/framework/skipper" + "k8s.io/kubernetes/test/e2e/storage/drivers" + "k8s.io/kubernetes/test/e2e/storage/utils" + admissionapi "k8s.io/pod-security-admission/api" +) + +var _ = utils.SIGDescribe("CSI Mock when kubelet restart", framework.WithSerial(), framework.WithDisruptive(), func() { + f := framework.NewDefaultFramework("csi-mock-when-kubelet-restart") + f.NamespacePodSecurityLevel = admissionapi.LevelPrivileged + m := newMockDriverSetup(f) + + ginkgo.BeforeEach(func() { + // These tests requires SSH to nodes, so the provider check should be identical to there + // (the limiting factor is the implementation of util.go's e2essh.GetSigner(...)). + + // Cluster must support node reboot + e2eskipper.SkipUnlessProviderIs(framework.ProvidersWithSSH...) + e2eskipper.SkipUnlessSSHKeyPresent() + }) + + ginkgo.It("should not umount volume when the pvc is terminating but still used by a running pod", func(ctx context.Context) { + + m.init(ctx, testParameters{ + registerDriver: true, + }) + ginkgo.DeferCleanup(m.cleanup) + + ginkgo.By("Creating a Pod with a PVC backed by a CSI volume") + _, pvc, pod := m.createPod(ctx, pvcReference) + + ginkgo.By("Waiting for the Pod to be running") + err := e2epod.WaitForPodRunningInNamespace(ctx, f.ClientSet, pod) + framework.ExpectNoError(err, "failed to wait for pod %s to be running", pod.Name) + pod, err = f.ClientSet.CoreV1().Pods(pod.Namespace).Get(ctx, pod.Name, metav1.GetOptions{}) + framework.ExpectNoError(err, "failed to get pod %s", pod.Name) + + ginkgo.By("Deleting the PVC") + err = f.ClientSet.CoreV1().PersistentVolumeClaims(pvc.Namespace).Delete(ctx, pvc.Name, metav1.DeleteOptions{}) + framework.ExpectNoError(err, "failed to delete PVC %s", pvc.Name) + + ginkgo.By("Restarting kubelet") + utils.KubeletCommand(ctx, utils.KRestart, f.ClientSet, pod) + ginkgo.DeferCleanup(utils.KubeletCommand, utils.KStart, f.ClientSet, pod) + + ginkgo.By("Verifying the PVC is terminating during kubelet restart") + pvc, err = f.ClientSet.CoreV1().PersistentVolumeClaims(pvc.Namespace).Get(ctx, pvc.Name, metav1.GetOptions{}) + framework.ExpectNoError(err, "failed to get PVC %s", pvc.Name) + gomega.Expect(pvc.DeletionTimestamp).NotTo(gomega.BeNil(), "PVC %s should have deletion timestamp", pvc.Name) + + ginkgo.By(fmt.Sprintf("Verifying that the driver didn't receive NodeUnpublishVolume call for PVC %s", pvc.Name)) + gomega.Consistently(ctx, + func(ctx context.Context) []drivers.MockCSICall { + calls, err := m.driver.GetCalls(ctx) + if err != nil { + if apierrors.IsUnexpectedServerError(err) { + // kubelet might not be ready yet when getting the calls + gomega.TryAgainAfter(framework.Poll).Wrap(err).Now() + return nil + } + return nil + } + return calls + }). + WithPolling(framework.Poll). + WithTimeout(framework.ClaimProvisionShortTimeout). + ShouldNot(gomega.ContainElement(gomega.HaveField("Method", gomega.Equal("NodeUnpublishVolume")))) + + ginkgo.By("Verifying the Pod is still running") + err = e2epod.WaitForPodRunningInNamespace(ctx, f.ClientSet, pod) + framework.ExpectNoError(err, "failed to wait for pod %s to be running", pod.Name) + }) +})