Retry recycle or delete operation on failure.

Recycle controller tries to recycle or delete a PV several times.
It stores count of failed attempts and timestamp of the last attempt in
annotations of the PV.

By default, the controller tries to recycle/delete a PV 3 times in
10 minutes interval. These values are configurable by
kube-controller-manager --pv-recycler-maximum-retry=X --pvclaimbinder-sync-period=Y
arguments.

cmd/kube-controller-manager/app/controllermanager.go

@@ -306,7 +306,9 @@ func StartControllers(s *options.CMServer, kubeClient *client.Client, kubeconfig
 
 	pvRecycler, err := persistentvolumecontroller.NewPersistentVolumeRecycler(
 		clientset.NewForConfigOrDie(client.AddUserAgent(kubeconfig, "persistent-volume-recycler")),
-		s.PersistentVolumeControllerOptions.PVClaimBinderSyncPeriod, ProbeRecyclableVolumePlugins(s.PersistentVolumeControllerOptions.VolumeConfigFlags),
+		s.PersistentVolumeControllerOptions.PVClaimBinderSyncPeriod,
+		s.PersistentVolumeControllerOptions.VolumeConfigFlags.PersistentVolumeRecyclerMaximumRetry,
+		ProbeRecyclableVolumePlugins(s.PersistentVolumeControllerOptions.VolumeConfigFlags),
 		cloud,
 	)
 	if err != nil {

contrib/mesos/pkg/controllermanager/controllermanager.go

@@ -256,6 +256,7 @@ func (s *CMServer) Run(_ []string) error {
 	pvRecycler, err := persistentvolumecontroller.NewPersistentVolumeRecycler(
 		clientset.NewForConfigOrDie(client.AddUserAgent(kubeconfig, "persistent-volume-recycler")),
 		s.PersistentVolumeControllerOptions.PVClaimBinderSyncPeriod,
+		s.PersistentVolumeControllerOptions.VolumeConfigFlags.PersistentVolumeRecyclerMaximumRetry,
 		kubecontrollermanager.ProbeRecyclableVolumePlugins(s.PersistentVolumeControllerOptions.VolumeConfigFlags), cloud)
 	if err != nil {
 		glog.Fatalf("Failed to start persistent volume recycler: %+v", err)

docs/admin/kube-controller-manager.md

@@ -91,6 +91,7 @@ kube-controller-manager
       --port=10252: The port that the controller-manager's http service runs on
       --profiling[=true]: Enable profiling via web interface host:port/debug/pprof/
       --pv-recycler-increment-timeout-nfs=30: the increment of time added per Gi to ActiveDeadlineSeconds for an NFS scrubber pod
+      --pv-recycler-maximum-retry=3: Maximum number of attempts to recycle or delete a persistent volume
       --pv-recycler-minimum-timeout-hostpath=60: The minimum ActiveDeadlineSeconds to use for a HostPath Recycler pod. This is for development and testing only and will not work in a multi-node cluster.
       --pv-recycler-minimum-timeout-nfs=300: The minimum ActiveDeadlineSeconds to use for an NFS Recycler pod
       --pv-recycler-pod-template-filepath-hostpath="": The file path to a pod definition used as a template for HostPath persistent volume recycling. This is for development and testing only and will not work in a multi-node cluster.

hack/verify-flags/known-flags.txt

@@ -279,6 +279,7 @@ pv-recycler-minimum-timeout-hostpath
 pv-recycler-minimum-timeout-nfs
 pv-recycler-pod-template-filepath-hostpath
 pv-recycler-pod-template-filepath-nfs
+pv-recycler-maximum-retry
 pv-recycler-timeout-increment-hostpath
 pvclaimbinder-sync-period
 read-only-port

pkg/controller/persistentvolume/options/options.go

@@ -27,6 +27,7 @@ import (
 // of volume.VolumeConfig which are then passed to the appropriate plugin. The ControllerManager binary is the only
 // part of the code which knows what plugins are supported and which CLI flags correspond to each plugin.
 type VolumeConfigFlags struct {
+	PersistentVolumeRecyclerMaximumRetry                int
 	PersistentVolumeRecyclerMinimumTimeoutNFS           int
 	PersistentVolumeRecyclerPodTemplateFilePathNFS      string
 	PersistentVolumeRecyclerIncrementTimeoutNFS         int
@@ -46,6 +47,7 @@ func NewPersistentVolumeControllerOptions() PersistentVolumeControllerOptions {
 		PVClaimBinderSyncPeriod: 10 * time.Minute,
 		VolumeConfigFlags: VolumeConfigFlags{
 			// default values here
+			PersistentVolumeRecyclerMaximumRetry:             3,
 			PersistentVolumeRecyclerMinimumTimeoutNFS:        300,
 			PersistentVolumeRecyclerIncrementTimeoutNFS:      30,
 			PersistentVolumeRecyclerMinimumTimeoutHostPath:   60,
@@ -76,6 +78,9 @@ func (o *PersistentVolumeControllerOptions) AddFlags(fs *pflag.FlagSet) {
 		o.VolumeConfigFlags.PersistentVolumeRecyclerIncrementTimeoutHostPath,
 		"the increment of time added per Gi to ActiveDeadlineSeconds for a HostPath scrubber pod. "+
 			"This is for development and testing only and will not work in a multi-node cluster.")
+	fs.IntVar(&o.VolumeConfigFlags.PersistentVolumeRecyclerMaximumRetry, "pv-recycler-maximum-retry",
+		o.VolumeConfigFlags.PersistentVolumeRecyclerMaximumRetry,
+		"Maximum number of attempts to recycle or delete a persistent volume")
 	fs.BoolVar(&o.VolumeConfigFlags.EnableHostPathProvisioning, "enable-hostpath-provisioner", o.VolumeConfigFlags.EnableHostPathProvisioning,
 		"Enable HostPath PV provisioning when running without a cloud provider. This allows testing and development of provisioning features. "+
 			"HostPath provisioning is not supported in any way, won't work in a multi-node cluster, and should not be used for anything other than testing or development.")

pkg/controller/persistentvolume/persistentvolume_recycler_controller.go

@@ -46,15 +46,33 @@ type PersistentVolumeRecycler struct {
 	kubeClient       clientset.Interface
 	pluginMgr        volume.VolumePluginMgr
 	cloud            cloudprovider.Interface
+	maximumRetry     int
+	syncPeriod       time.Duration
+	// Local cache of failed recycle / delete operations. Map volume.Name -> status of the volume.
+	// Only PVs in Released state have an entry here.
+	releasedVolumes map[string]releasedVolumeStatus
 }
 
-// PersistentVolumeRecycler creates a new PersistentVolumeRecycler
-func NewPersistentVolumeRecycler(kubeClient clientset.Interface, syncPeriod time.Duration, plugins []volume.VolumePlugin, cloud cloudprovider.Interface) (*PersistentVolumeRecycler, error) {
+// releasedVolumeStatus holds state of failed delete/recycle operation on a
+// volume. The controller re-tries the operation several times and it stores
+// retry count + timestamp of the last attempt here.
+type releasedVolumeStatus struct {
+	// How many recycle/delete operations failed.
+	retryCount int
+	// Timestamp of the last attempt.
+	lastAttempt time.Time
+}
+
+// NewPersistentVolumeRecycler creates a new PersistentVolumeRecycler
+func NewPersistentVolumeRecycler(kubeClient clientset.Interface, syncPeriod time.Duration, maximumRetry int, plugins []volume.VolumePlugin, cloud cloudprovider.Interface) (*PersistentVolumeRecycler, error) {
 	recyclerClient := NewRecyclerClient(kubeClient)
 	recycler := &PersistentVolumeRecycler{
-		client:     recyclerClient,
-		kubeClient: kubeClient,
-		cloud:      cloud,
+		client:          recyclerClient,
+		kubeClient:      kubeClient,
+		cloud:           cloud,
+		maximumRetry:    maximumRetry,
+		syncPeriod:      syncPeriod,
+		releasedVolumes: make(map[string]releasedVolumeStatus),
 	}
 
 	if err := recycler.pluginMgr.InitPlugins(plugins, recycler); err != nil {
@@ -89,6 +107,14 @@ func NewPersistentVolumeRecycler(kubeClient clientset.Interface, syncPeriod time
 				}
 				recycler.reclaimVolume(pv)
 			},
+			DeleteFunc: func(obj interface{}) {
+				pv, ok := obj.(*api.PersistentVolume)
+				if !ok {
+					glog.Errorf("Error casting object to PersistentVolume: %v", obj)
+					return
+				}
+				recycler.removeReleasedVolume(pv)
+			},
 		},
 	)
 
@@ -96,17 +122,50 @@ func NewPersistentVolumeRecycler(kubeClient clientset.Interface, syncPeriod time
 	return recycler, nil
 }
 
-func (recycler *PersistentVolumeRecycler) reclaimVolume(pv *api.PersistentVolume) error {
-	if pv.Status.Phase == api.VolumeReleased && pv.Spec.ClaimRef != nil {
-		glog.V(5).Infof("Reclaiming PersistentVolume[%s]\n", pv.Name)
+// shouldRecycle checks a volume and returns nil, if the volume should be
+// recycled right now. Otherwise it returns an error with reason why it should
+// not be recycled.
+func (recycler *PersistentVolumeRecycler) shouldRecycle(pv *api.PersistentVolume) error {
+	if pv.Spec.ClaimRef == nil {
+		return fmt.Errorf("Volume does not have a reference to claim")
+	}
+	if pv.Status.Phase != api.VolumeReleased {
+		return fmt.Errorf("The volume is not in 'Released' phase")
+	}
 
-		latest, err := recycler.client.GetPersistentVolume(pv.Name)
-		if err != nil {
-			return fmt.Errorf("Could not find PersistentVolume %s", pv.Name)
-		}
-		if latest.Status.Phase != api.VolumeReleased {
-			return fmt.Errorf("PersistentVolume[%s] phase is %s, expected %s.  Skipping.", pv.Name, latest.Status.Phase, api.VolumeReleased)
-		}
+	// The volume is Released, should we retry recycling?
+	status, found := recycler.releasedVolumes[pv.Name]
+	if !found {
+		// We don't know anything about this volume. The controller has been
+		// restarted or the volume has been marked as Released by another
+		// controller. Recycle/delete this volume as if it was just Released.
+		glog.V(5).Infof("PersistentVolume[%s] not found in local cache, recycling", pv.Name)
+		return nil
+	}
+
+	// Check the timestamp
+	expectedRetry := status.lastAttempt.Add(recycler.syncPeriod)
+	if time.Now().After(expectedRetry) {
+		glog.V(5).Infof("PersistentVolume[%s] retrying recycle after timeout", pv.Name)
+		return nil
+	}
+	// It's too early
+	glog.V(5).Infof("PersistentVolume[%s] skipping recycle, it's too early: now: %v, next retry: %v", pv.Name, time.Now(), expectedRetry)
+	return fmt.Errorf("Too early after previous failure")
+}
+
+func (recycler *PersistentVolumeRecycler) reclaimVolume(pv *api.PersistentVolume) error {
+	glog.V(5).Infof("Recycler: checking PersistentVolume[%s]\n", pv.Name)
+	// Always load the latest version of the volume
+	newPV, err := recycler.client.GetPersistentVolume(pv.Name)
+	if err != nil {
+		return fmt.Errorf("Could not find PersistentVolume %s", pv.Name)
+	}
+	pv = newPV
+
+	err = recycler.shouldRecycle(pv)
+	if err == nil {
+		glog.V(5).Infof("Reclaiming PersistentVolume[%s]\n", pv.Name)
 
 		// both handleRecycle and handleDelete block until completion
 		// TODO: allow parallel recycling operations to increase throughput
@@ -125,10 +184,41 @@ func (recycler *PersistentVolumeRecycler) reclaimVolume(pv *api.PersistentVolume
 			glog.Errorf(errMsg)
 			return fmt.Errorf(errMsg)
 		}
+		return nil
 	}
+	glog.V(3).Infof("PersistentVolume[%s] phase %s - skipping: %v", pv.Name, pv.Status.Phase, err)
 	return nil
 }
 
+// handleReleaseFailure evaluates a failed Recycle/Delete operation, updates
+// internal controller state with new nr. of attempts and timestamp of the last
+// attempt. Based on the number of failures it returns the next state of the
+// volume (Released / Failed).
+func (recycler *PersistentVolumeRecycler) handleReleaseFailure(pv *api.PersistentVolume) api.PersistentVolumePhase {
+	status, found := recycler.releasedVolumes[pv.Name]
+	if !found {
+		// First failure, set retryCount to 0 (will be inceremented few lines below)
+		status = releasedVolumeStatus{}
+	}
+	status.retryCount += 1
+
+	if status.retryCount > recycler.maximumRetry {
+		// This was the last attempt. Remove any internal state and mark the
+		// volume as Failed.
+		glog.V(3).Infof("PersistentVolume[%s] failed %d times - marking Failed", pv.Name, status.retryCount)
+		recycler.removeReleasedVolume(pv)
+		return api.VolumeFailed
+	}
+
+	status.lastAttempt = time.Now()
+	recycler.releasedVolumes[pv.Name] = status
+	return api.VolumeReleased
+}
+
+func (recycler *PersistentVolumeRecycler) removeReleasedVolume(pv *api.PersistentVolume) {
+	delete(recycler.releasedVolumes, pv.Name)
+}
+
 func (recycler *PersistentVolumeRecycler) handleRecycle(pv *api.PersistentVolume) error {
 	glog.V(5).Infof("Recycling PersistentVolume[%s]\n", pv.Name)
 
@@ -154,9 +244,12 @@ func (recycler *PersistentVolumeRecycler) handleRecycle(pv *api.PersistentVolume
 		if err := volRecycler.Recycle(); err != nil {
 			glog.Errorf("PersistentVolume[%s] failed recycling: %+v", pv.Name, err)
 			pv.Status.Message = fmt.Sprintf("Recycling error: %s", err)
-			nextPhase = api.VolumeFailed
+			nextPhase = recycler.handleReleaseFailure(pv)
 		} else {
 			glog.V(5).Infof("PersistentVolume[%s] successfully recycled\n", pv.Name)
+			// The volume has been recycled. Remove any internal state to make
+			// any subsequent bind+recycle cycle working.
+			recycler.removeReleasedVolume(pv)
 			nextPhase = api.VolumePending
 		}
 	}
@@ -200,9 +293,10 @@ func (recycler *PersistentVolumeRecycler) handleDelete(pv *api.PersistentVolume)
 		if err != nil {
 			glog.Errorf("PersistentVolume[%s] failed deletion: %+v", pv.Name, err)
 			pv.Status.Message = fmt.Sprintf("Deletion error: %s", err)
-			nextPhase = api.VolumeFailed
+			nextPhase = recycler.handleReleaseFailure(pv)
 		} else {
 			glog.V(5).Infof("PersistentVolume[%s] successfully deleted through plugin\n", pv.Name)
+			recycler.removeReleasedVolume(pv)
 			// after successful deletion through the plugin, we can also remove the PV from the cluster
 			if err := recycler.client.DeletePersistentVolume(pv); err != nil {
 				return fmt.Errorf("error deleting persistent volume: %+v", err)

pkg/controller/persistentvolume/persistentvolume_recycler_controller_test.go

@@ -17,16 +17,149 @@ limitations under the License.
 package persistentvolume
 
 import (
+	"fmt"
 	"testing"
+	"time"
 
 	"k8s.io/kubernetes/pkg/api"
 	"k8s.io/kubernetes/pkg/api/resource"
 	"k8s.io/kubernetes/pkg/client/testing/fake"
 	"k8s.io/kubernetes/pkg/volume"
+	"k8s.io/kubernetes/pkg/volume/host_path"
+)
+
+const (
+	mySyncPeriod   = 2 * time.Second
+	myMaximumRetry = 3
 )
 
 func TestFailedRecycling(t *testing.T) {
-	pv := &api.PersistentVolume{
+	pv := preparePV()
+
+	mockClient := &mockBinderClient{
+		volume: pv,
+	}
+
+	// no Init called for pluginMgr and no plugins are available.  Volume should fail recycling.
+	plugMgr := volume.VolumePluginMgr{}
+
+	recycler := &PersistentVolumeRecycler{
+		kubeClient:      fake.NewSimpleClientset(),
+		client:          mockClient,
+		pluginMgr:       plugMgr,
+		releasedVolumes: make(map[string]releasedVolumeStatus),
+	}
+
+	err := recycler.reclaimVolume(pv)
+	if err != nil {
+		t.Errorf("Unexpected non-nil error: %v", err)
+	}
+
+	if mockClient.volume.Status.Phase != api.VolumeFailed {
+		t.Errorf("Expected %s but got %s", api.VolumeFailed, mockClient.volume.Status.Phase)
+	}
+
+	// Use a new volume for the next test
+	pv = preparePV()
+	mockClient.volume = pv
+
+	pv.Spec.PersistentVolumeReclaimPolicy = api.PersistentVolumeReclaimDelete
+	err = recycler.reclaimVolume(pv)
+	if err != nil {
+		t.Errorf("Unexpected non-nil error: %v", err)
+	}
+
+	if mockClient.volume.Status.Phase != api.VolumeFailed {
+		t.Errorf("Expected %s but got %s", api.VolumeFailed, mockClient.volume.Status.Phase)
+	}
+}
+
+func TestRecyclingRetry(t *testing.T) {
+	// Test that recycler controller retries to recycle a volume several times, which succeeds eventually
+	pv := preparePV()
+
+	mockClient := &mockBinderClient{
+		volume: pv,
+	}
+
+	plugMgr := volume.VolumePluginMgr{}
+	// Use a fake NewRecycler function
+	plugMgr.InitPlugins(host_path.ProbeRecyclableVolumePlugins(newFailingMockRecycler, volume.VolumeConfig{}), volume.NewFakeVolumeHost("/tmp/fake", nil, nil))
+	// Reset a global call counter
+	failedCallCount = 0
+
+	recycler := &PersistentVolumeRecycler{
+		kubeClient:      fake.NewSimpleClientset(),
+		client:          mockClient,
+		pluginMgr:       plugMgr,
+		syncPeriod:      mySyncPeriod,
+		maximumRetry:    myMaximumRetry,
+		releasedVolumes: make(map[string]releasedVolumeStatus),
+	}
+
+	// All but the last attempt will fail
+	testRecycleFailures(t, recycler, mockClient, pv, myMaximumRetry-1)
+
+	// The last attempt should succeed
+	err := recycler.reclaimVolume(pv)
+	if err != nil {
+		t.Errorf("Last step: Recycler failed: %v", err)
+	}
+
+	if mockClient.volume.Status.Phase != api.VolumePending {
+		t.Errorf("Last step: The volume should be Pending, but is %s instead", mockClient.volume.Status.Phase)
+	}
+	// Check the cache, it should not have any entry
+	status, found := recycler.releasedVolumes[pv.Name]
+	if found {
+		t.Errorf("Last step: Expected PV to be removed from cache, got %v", status)
+	}
+}
+
+func TestRecyclingRetryAlwaysFail(t *testing.T) {
+	// Test that recycler controller retries to recycle a volume several times, which always fails.
+	pv := preparePV()
+
+	mockClient := &mockBinderClient{
+		volume: pv,
+	}
+
+	plugMgr := volume.VolumePluginMgr{}
+	// Use a fake NewRecycler function
+	plugMgr.InitPlugins(host_path.ProbeRecyclableVolumePlugins(newAlwaysFailingMockRecycler, volume.VolumeConfig{}), volume.NewFakeVolumeHost("/tmp/fake", nil, nil))
+	// Reset a global call counter
+	failedCallCount = 0
+
+	recycler := &PersistentVolumeRecycler{
+		kubeClient:      fake.NewSimpleClientset(),
+		client:          mockClient,
+		pluginMgr:       plugMgr,
+		syncPeriod:      mySyncPeriod,
+		maximumRetry:    myMaximumRetry,
+		releasedVolumes: make(map[string]releasedVolumeStatus),
+	}
+
+	// myMaximumRetry recycle attempts will fail
+	testRecycleFailures(t, recycler, mockClient, pv, myMaximumRetry)
+
+	// The volume should be failed after myMaximumRetry attempts
+	err := recycler.reclaimVolume(pv)
+	if err != nil {
+		t.Errorf("Last step: Recycler failed: %v", err)
+	}
+
+	if mockClient.volume.Status.Phase != api.VolumeFailed {
+		t.Errorf("Last step: The volume should be Failed, but is %s instead", mockClient.volume.Status.Phase)
+	}
+	// Check the cache, it should not have any entry
+	status, found := recycler.releasedVolumes[pv.Name]
+	if found {
+		t.Errorf("Last step: Expected PV to be removed from cache, got %v", status)
+	}
+}
+
+func preparePV() *api.PersistentVolume {
+	return &api.PersistentVolume{
 		Spec: api.PersistentVolumeSpec{
 			AccessModes: []api.PersistentVolumeAccessMode{api.ReadWriteOnce},
 			Capacity: api.ResourceList{
@@ -34,7 +167,7 @@ func TestFailedRecycling(t *testing.T) {
 			},
 			PersistentVolumeSource: api.PersistentVolumeSource{
 				HostPath: &api.HostPathVolumeSource{
-					Path: "/somepath/data02",
+					Path: "/tmp/data02",
 				},
 			},
 			PersistentVolumeReclaimPolicy: api.PersistentVolumeReclaimRecycle,
@@ -47,36 +180,85 @@ func TestFailedRecycling(t *testing.T) {
 			Phase: api.VolumeReleased,
 		},
 	}
+}
 
-	mockClient := &mockBinderClient{
-		volume: pv,
-	}
+// Test that `count` attempts to recycle a PV fails.
+func testRecycleFailures(t *testing.T, recycler *PersistentVolumeRecycler, mockClient *mockBinderClient, pv *api.PersistentVolume, count int) {
+	for i := 1; i <= count; i++ {
+		err := recycler.reclaimVolume(pv)
+		if err != nil {
+			t.Errorf("STEP %d: Recycler faled: %v", i, err)
+		}
 
-	// no Init called for pluginMgr and no plugins are available.  Volume should fail recycling.
-	plugMgr := volume.VolumePluginMgr{}
+		// Check the status, it should be failed
+		if mockClient.volume.Status.Phase != api.VolumeReleased {
+			t.Errorf("STEP %d: The volume should be Released, but is %s instead", i, mockClient.volume.Status.Phase)
+		}
 
-	recycler := &PersistentVolumeRecycler{
-		kubeClient: fake.NewSimpleClientset(),
-		client:     mockClient,
-		pluginMgr:  plugMgr,
-	}
+		// Check the failed volume cache
+		status, found := recycler.releasedVolumes[pv.Name]
+		if !found {
+			t.Errorf("STEP %d: cannot find released volume status", i)
+		}
+		if status.retryCount != i {
+			t.Errorf("STEP %d: Expected nr. of attempts to be %d, got %d", i, i, status.retryCount)
+		}
 
-	err := recycler.reclaimVolume(pv)
-	if err != nil {
-		t.Errorf("Unexpected non-nil error: %v", err)
-	}
+		// call reclaimVolume too early, it should not increment the retryCount
+		time.Sleep(mySyncPeriod / 2)
+		err = recycler.reclaimVolume(pv)
+		if err != nil {
+			t.Errorf("STEP %d: Recycler failed: %v", i, err)
+		}
 
-	if mockClient.volume.Status.Phase != api.VolumeFailed {
-		t.Errorf("Expected %s but got %s", api.VolumeFailed, mockClient.volume.Status.Phase)
-	}
+		status, found = recycler.releasedVolumes[pv.Name]
+		if !found {
+			t.Errorf("STEP %d: cannot find released volume status", i)
+		}
+		if status.retryCount != i {
+			t.Errorf("STEP %d: Expected nr. of attempts to be %d, got %d", i, i, status.retryCount)
+		}
 
-	pv.Spec.PersistentVolumeReclaimPolicy = api.PersistentVolumeReclaimDelete
-	err = recycler.reclaimVolume(pv)
-	if err != nil {
-		t.Errorf("Unexpected non-nil error: %v", err)
-	}
-
-	if mockClient.volume.Status.Phase != api.VolumeFailed {
-		t.Errorf("Expected %s but got %s", api.VolumeFailed, mockClient.volume.Status.Phase)
+		// Call the next reclaimVolume() after full pvRecycleRetryPeriod
+		time.Sleep(mySyncPeriod / 2)
 	}
 }
+
+func newFailingMockRecycler(spec *volume.Spec, host volume.VolumeHost, config volume.VolumeConfig) (volume.Recycler, error) {
+	return &failingMockRecycler{
+		path:       spec.PersistentVolume.Spec.HostPath.Path,
+		errorCount: myMaximumRetry - 1, // fail two times and then successfuly recycle the volume
+	}, nil
+}
+
+func newAlwaysFailingMockRecycler(spec *volume.Spec, host volume.VolumeHost, config volume.VolumeConfig) (volume.Recycler, error) {
+	return &failingMockRecycler{
+		path:       spec.PersistentVolume.Spec.HostPath.Path,
+		errorCount: 1000, // always fail
+	}, nil
+}
+
+type failingMockRecycler struct {
+	path string
+	// How many times should the recycler fail before returning success.
+	errorCount int
+	volume.MetricsNil
+}
+
+// Counter of failingMockRecycler.Recycle() calls. Global variable just for
+// testing. It's too much code to create a custom volume plugin, which would
+// hold this variable.
+var failedCallCount = 0
+
+func (r *failingMockRecycler) GetPath() string {
+	return r.path
+}
+
+func (r *failingMockRecycler) Recycle() error {
+	failedCallCount += 1
+	if failedCallCount <= r.errorCount {
+		return fmt.Errorf("Failing for %d. time", failedCallCount)
+	}
+	// return nil means recycle passed
+	return nil
+}

test/integration/persistent_volumes_test.go

@@ -58,7 +58,7 @@ func TestPersistentVolumeRecycler(t *testing.T) {
 	binder.Run()
 	defer binder.Stop()
 
-	recycler, _ := persistentvolumecontroller.NewPersistentVolumeRecycler(recyclerClient, 30*time.Second, plugins, cloud)
+	recycler, _ := persistentvolumecontroller.NewPersistentVolumeRecycler(recyclerClient, 30*time.Second, 3, plugins, cloud)
 	recycler.Run()
 	defer recycler.Stop()