reclaim terminated pod volumes

cmd/kubelet/app/options/options.go

@@ -251,6 +251,7 @@ func (s *KubeletServer) AddFlags(fs *pflag.FlagSet) {
 	fs.BoolVar(&s.ExperimentalKernelMemcgNotification, "experimental-kernel-memcg-notification", s.ExperimentalKernelMemcgNotification, "If enabled, the kubelet will integrate with the kernel memcg notification to determine if memory eviction thresholds are crossed rather than polling.")
 	fs.Int32Var(&s.PodsPerCore, "pods-per-core", s.PodsPerCore, "Number of Pods per core that can run on this Kubelet. The total number of Pods on this Kubelet cannot exceed max-pods, so max-pods will be used if this calculation results in a larger number of Pods allowed on the Kubelet. A value of 0 disables this limit.")
 	fs.BoolVar(&s.ProtectKernelDefaults, "protect-kernel-defaults", s.ProtectKernelDefaults, "Default kubelet behaviour for kernel tuning. If set, kubelet errors if any of kernel tunables is different than kubelet defaults.")
+	fs.BoolVar(&s.KeepTerminatedPodVolumes, "keep-terminated-pod-volumes", s.KeepTerminatedPodVolumes, "Keep terminated pod volumes mounted to the node after the pod terminates.  Can be useful for debugging volume related issues.")
 
 	// CRI flags.
 	fs.BoolVar(&s.EnableCRI, "experimental-cri", s.EnableCRI, "[Experimental] Enable the Container Runtime Interface (CRI) integration. If --container-runtime is set to \"remote\", Kubelet will communicate with the runtime/image CRI server listening on the endpoint specified by --remote-runtime-endpoint/--remote-image-endpoint. If --container-runtime is set to \"docker\", Kubelet will launch a in-process CRI server on behalf of docker, and communicate over a default endpoint.")

hack/local-up-cluster.sh

@@ -536,6 +536,7 @@ function start_kubelet {
         --experimental-cgroups-per-qos=${EXPERIMENTAL_CGROUPS_PER_QOS} \
         --cgroup-driver=${CGROUP_DRIVER} \
         --cgroup-root=${CGROUP_ROOT} \
+        --keep-terminated-pod-volumes=true \
         ${auth_args} \
         ${dns_args} \
         ${net_plugin_dir_args} \

hack/verify-flags/known-flags.txt

@@ -308,6 +308,7 @@ junit-file-number
 k8s-bin-dir
 k8s-build-output
 keep-gogoproto
+keep-terminated-pod-volumes
 km-path
 kops-admin-access
 kops-cluster

pkg/apis/componentconfig/types.go

@@ -481,6 +481,9 @@ type KubeletConfiguration struct {
 	// (binaries, etc.) to mount the volume are available on the underlying node. If the check is enabled
 	// and fails the mount operation fails.
 	ExperimentalCheckNodeCapabilitiesBeforeMount bool
+	// This flag, if set, instructs the kubelet to keep volumes from terminated pods mounted to the node.
+	// This can be useful for debugging volume related issues.
+	KeepTerminatedPodVolumes bool
 }
 
 type KubeletAuthorizationMode string

pkg/apis/componentconfig/v1alpha1/types.go

@@ -521,6 +521,9 @@ type KubeletConfiguration struct {
 	// (binaries, etc.) to mount the volume are available on the underlying node. If the check is enabled
 	// and fails the mount operation fails.
 	ExperimentalCheckNodeCapabilitiesBeforeMount bool `json:"experimentalCheckNodeCapabilitiesBeforeMount,omitempty"`
+	// This flag, if set, instructs the kubelet to keep volumes from terminated pods mounted to the node.
+	// This can be useful for debugging volume related issues.
+	KeepTerminatedPodVolumes bool `json:"keepTerminatedPodVolumes,omitempty"`
 }
 
 type KubeletAuthorizationMode string

pkg/apis/componentconfig/v1alpha1/zz_generated.conversion.go

@@ -481,6 +481,7 @@ func autoConvert_v1alpha1_KubeletConfiguration_To_componentconfig_KubeletConfigu
 	out.EnableCRI = in.EnableCRI
 	out.ExperimentalFailSwapOn = in.ExperimentalFailSwapOn
 	out.ExperimentalCheckNodeCapabilitiesBeforeMount = in.ExperimentalCheckNodeCapabilitiesBeforeMount
+	out.KeepTerminatedPodVolumes = in.KeepTerminatedPodVolumes
 	return nil
 }
 
@@ -652,6 +653,7 @@ func autoConvert_componentconfig_KubeletConfiguration_To_v1alpha1_KubeletConfigu
 	out.EnableCRI = in.EnableCRI
 	out.ExperimentalFailSwapOn = in.ExperimentalFailSwapOn
 	out.ExperimentalCheckNodeCapabilitiesBeforeMount = in.ExperimentalCheckNodeCapabilitiesBeforeMount
+	out.KeepTerminatedPodVolumes = in.KeepTerminatedPodVolumes
 	return nil
 }
 

pkg/generated/openapi/zz_generated.openapi.go

@@ -10431,6 +10431,13 @@ var OpenAPIDefinitions *openapi.OpenAPIDefinitions = &openapi.OpenAPIDefinitions
 							Format:      "",
 						},
 					},
+					"keepTerminatedPodVolumes": {
+						SchemaProps: spec.SchemaProps{
+							Description: "This flag, if set, instructs the kubelet to keep volumes from terminated pods mounted to the node. This can be useful for debugging volume related issues.",
+							Type:        []string{"boolean"},
+							Format:      "",
+						},
+					},
 				},
 				Required: []string{"podManifestPath", "syncFrequency", "fileCheckFrequency", "httpCheckFrequency", "manifestURL", "manifestURLHeader", "enableServer", "address", "port", "readOnlyPort", "tlsCertFile", "tlsPrivateKeyFile", "certDirectory", "authentication", "authorization", "hostnameOverride", "podInfraContainerImage", "dockerEndpoint", "rootDirectory", "seccompProfileRoot", "allowPrivileged", "hostNetworkSources", "hostPIDSources", "hostIPCSources", "registryPullQPS", "registryBurst", "eventRecordQPS", "eventBurst", "enableDebuggingHandlers", "minimumGCAge", "maxPerPodContainerCount", "maxContainerCount", "cAdvisorPort", "healthzPort", "healthzBindAddress", "oomScoreAdj", "registerNode", "clusterDomain", "masterServiceNamespace", "clusterDNS", "streamingConnectionIdleTimeout", "nodeStatusUpdateFrequency", "imageMinimumGCAge", "imageGCHighThresholdPercent", "imageGCLowThresholdPercent", "lowDiskSpaceThresholdMB", "volumeStatsAggPeriod", "networkPluginName", "networkPluginDir", "cniConfDir", "cniBinDir", "networkPluginMTU", "volumePluginDir", "cloudProvider", "cloudConfigFile", "kubeletCgroups", "runtimeCgroups", "systemCgroups", "cgroupRoot", "containerRuntime", "remoteRuntimeEndpoint", "remoteImageEndpoint", "runtimeRequestTimeout", "rktPath", "rktAPIEndpoint", "rktStage1Image", "lockFilePath", "exitOnLockContention", "hairpinMode", "babysitDaemons", "maxPods", "nvidiaGPUs", "dockerExecHandlerName", "podCIDR", "resolvConf", "cpuCFSQuota", "containerized", "maxOpenFiles", "registerSchedulable", "registerWithTaints", "contentType", "kubeAPIQPS", "kubeAPIBurst", "serializeImagePulls", "outOfDiskTransitionFrequency", "nodeIP", "nodeLabels", "nonMasqueradeCIDR", "enableCustomMetrics", "evictionHard", "evictionSoft", "evictionSoftGracePeriod", "evictionPressureTransitionPeriod", "evictionMaxPodGracePeriod", "evictionMinimumReclaim", "experimentalKernelMemcgNotification", "podsPerCore", "enableControllerAttachDetach", "systemReserved", "kubeReserved", "protectKernelDefaults", "makeIPTablesUtilChains", "iptablesMasqueradeBit", "iptablesDropBit"},
 			},

pkg/kubelet/kubelet.go

@@ -747,7 +747,8 @@ func NewMainKubelet(kubeCfg *componentconfig.KubeletConfiguration, kubeDeps *Kub
 		kubeDeps.Mounter,
 		klet.getPodsDir(),
 		kubeDeps.Recorder,
-		kubeCfg.ExperimentalCheckNodeCapabilitiesBeforeMount)
+		kubeCfg.ExperimentalCheckNodeCapabilitiesBeforeMount,
+		kubeCfg.KeepTerminatedPodVolumes)
 
 	runtimeCache, err := kubecontainer.NewRuntimeCache(klet.containerRuntime)
 	if err != nil {

pkg/kubelet/kubelet_test.go

@@ -257,7 +257,8 @@ func newTestKubeletWithImageList(
 		kubelet.mounter,
 		kubelet.getPodsDir(),
 		kubelet.recorder,
-		false /* experimentalCheckNodeCapabilitiesBeforeMount*/)
+		false, /* experimentalCheckNodeCapabilitiesBeforeMount*/
+		false /* keepTerminatedPodVolumes */)
 	require.NoError(t, err, "Failed to initialize volume manager")
 
 	// enable active deadline handler

pkg/kubelet/runonce_test.go

@@ -106,7 +106,8 @@ func TestRunOnce(t *testing.T) {
 		kb.mounter,
 		kb.getPodsDir(),
 		kb.recorder,
-		false /* experimentalCheckNodeCapabilitiesBeforeMount */)
+		false, /* experimentalCheckNodeCapabilitiesBeforeMount */
+		false /* keepTerminatedPodVolumes */)
 
 	kb.networkPlugin, _ = network.InitNetworkPlugin([]network.NetworkPlugin{}, "", nettest.NewFakeHost(nil), componentconfig.HairpinNone, kb.nonMasqueradeCIDR, network.UseDefaultMTU)
 	// TODO: Factor out "StatsProvider" from Kubelet so we don't have a cyclic dependency

pkg/kubelet/volumemanager/populator/desired_state_of_world_populator.go

@@ -71,7 +71,8 @@ func NewDesiredStateOfWorldPopulator(
 	getPodStatusRetryDuration time.Duration,
 	podManager pod.Manager,
 	desiredStateOfWorld cache.DesiredStateOfWorld,
-	kubeContainerRuntime kubecontainer.Runtime) DesiredStateOfWorldPopulator {
+	kubeContainerRuntime kubecontainer.Runtime,
+	keepTerminatedPodVolumes bool) DesiredStateOfWorldPopulator {
 	return &desiredStateOfWorldPopulator{
 		kubeClient:                kubeClient,
 		loopSleepDuration:         loopSleepDuration,
@@ -80,7 +81,8 @@ func NewDesiredStateOfWorldPopulator(
 		desiredStateOfWorld:       desiredStateOfWorld,
 		pods: processedPods{
 			processedPods: make(map[volumetypes.UniquePodName]bool)},
-		kubeContainerRuntime: kubeContainerRuntime,
+		kubeContainerRuntime:     kubeContainerRuntime,
+		keepTerminatedPodVolumes: keepTerminatedPodVolumes,
 	}
 }
 
@@ -93,6 +95,7 @@ type desiredStateOfWorldPopulator struct {
 	pods                      processedPods
 	kubeContainerRuntime      kubecontainer.Runtime
 	timeOfLastGetPodStatus    time.Time
+	keepTerminatedPodVolumes  bool
 }
 
 type processedPods struct {
@@ -160,13 +163,7 @@ func (dswp *desiredStateOfWorldPopulator) findAndRemoveDeletedPods() {
 			if !isPodTerminated(pod) {
 				continue
 			}
-			// Skip non-memory backed volumes belonging to terminated pods
-			volume := volumeToMount.VolumeSpec.Volume
-			if volume == nil {
-				continue
-			}
-			if (volume.EmptyDir == nil || volume.EmptyDir.Medium != v1.StorageMediumMemory) &&
-				volume.ConfigMap == nil && volume.Secret == nil {
+			if dswp.keepTerminatedPodVolumes {
 				continue
 			}
 		}

pkg/kubelet/volumemanager/volume_manager.go

@@ -157,7 +157,8 @@ func NewVolumeManager(
 	mounter mount.Interface,
 	kubeletPodsDir string,
 	recorder record.EventRecorder,
-	checkNodeCapabilitiesBeforeMount bool) (VolumeManager, error) {
+	checkNodeCapabilitiesBeforeMount bool,
+	keepTerminatedPodVolumes bool) (VolumeManager, error) {
 
 	vm := &volumeManager{
 		kubeClient:          kubeClient,
@@ -191,7 +192,8 @@ func NewVolumeManager(
 		desiredStateOfWorldPopulatorGetPodStatusRetryDuration,
 		podManager,
 		vm.desiredStateOfWorld,
-		kubeContainerRuntime)
+		kubeContainerRuntime,
+		keepTerminatedPodVolumes)
 
 	return vm, nil
 }

pkg/kubelet/volumemanager/volume_manager_test.go

@@ -198,7 +198,8 @@ func newTestVolumeManager(
 		&mount.FakeMounter{},
 		"",
 		fakeRecorder,
-		false /* experimentalCheckNodeCapabilitiesBeforeMount */)
+		false, /* experimentalCheckNodeCapabilitiesBeforeMount */
+		false /* keepTerminatedPodVolumes */)
 
 	return vm, err
 }