update PVC regardless of whether the Pod is terminating

Signed-off-by: Abner-1 <yuanyuxing.yyx@alibaba-inc.com>

apis/apps/v1beta1/statefulset_types.go

@@ -171,7 +171,7 @@ const (
 )
 
 // StatefulSetPersistentVolumeClaimRetentionPolicy describes the policy used for PVCs
-// created from the StatefulSet VolumeClaimTemplates.
+// created from the StatefulSet VolumeClaims.
 type StatefulSetPersistentVolumeClaimRetentionPolicy struct {
 	// WhenDeleted specifies what happens to PVCs created from StatefulSet
 	// VolumeClaimTemplates when the StatefulSet is deleted. The default policy
@@ -365,11 +365,11 @@ type StatefulSetStatus struct {
 	// LabelSelector is label selectors for query over pods that should match the replica count used by HPA.
 	LabelSelector string `json:"labelSelector,omitempty"`
 
-	// VolumeClaimTemplates represents the status of compatibility between existing VolumeClaims
-	// and their respective templates. It tracks whether the VolumeClaims have been updated
-	// to match any changes made to the VolumeClaimTemplates, ensuring synchronization
+	// VolumeClaims represents the status of compatibility between existing PVCs
+	// and their respective templates. It tracks whether the PersistentVolumeClaims have been updated
+	// to match any changes made to the volumeClaimTemplates, ensuring synchronization
 	// between the defined templates and the actual PersistentVolumeClaims in use.
-	VolumeClaimTemplates []VolumeClaimStatus `json:"volumeClaimTemplates,omitempty"`
+	VolumeClaims []VolumeClaimStatus `json:"volumeClaims,omitempty"`
 }
 
 // These are valid conditions of a statefulset.

config/crd/bases/apps.kruise.io_statefulsets.yaml

@@ -1036,11 +1036,11 @@ spec:
                   indicated by updateRevision.
                 format: int32
                 type: integer
-              volumeClaimTemplates:
+              volumeClaims:
                 description: |-
-                  VolumeClaimTemplates represents the status of compatibility between existing VolumeClaims
-                  and their respective templates. It tracks whether the VolumeClaims have been updated
-                  to match any changes made to the VolumeClaimTemplates, ensuring synchronization
+                  VolumeClaims represents the status of compatibility between existing PVCs
+                  and their respective templates. It tracks whether the PersistentVolumeClaims have been updated
+                  to match any changes made to the volumeClaimTemplates, ensuring synchronization
                   between the defined templates and the actual PersistentVolumeClaims in use.
                 items:
                   description: |-

pkg/controller/cloneset/core/cloneset_core.go

@@ -22,16 +22,17 @@ import (
 	"regexp"
 
 	"github.com/appscode/jsonpatch"
-	appspub "github.com/openkruise/kruise/apis/apps/pub"
-	appsv1alpha1 "github.com/openkruise/kruise/apis/apps/v1alpha1"
-	clonesetutils "github.com/openkruise/kruise/pkg/controller/cloneset/utils"
-	"github.com/openkruise/kruise/pkg/util/inplaceupdate"
 	v1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/klog/v2"
 	podutil "k8s.io/kubernetes/pkg/api/v1/pod"
 	kubecontroller "k8s.io/kubernetes/pkg/controller"
 	"sigs.k8s.io/controller-runtime/pkg/controller/controllerutil"
+
+	appspub "github.com/openkruise/kruise/apis/apps/pub"
+	appsv1alpha1 "github.com/openkruise/kruise/apis/apps/v1alpha1"
+	clonesetutils "github.com/openkruise/kruise/pkg/controller/cloneset/utils"
+	"github.com/openkruise/kruise/pkg/util/inplaceupdate"
 )
 
 var (

pkg/controller/cloneset/utils/cloneset_utils.go

@@ -22,13 +22,6 @@ import (
 	"strings"
 	"sync"
 
-	appsv1alpha1 "github.com/openkruise/kruise/apis/apps/v1alpha1"
-	"github.com/openkruise/kruise/pkg/features"
-	utilclient "github.com/openkruise/kruise/pkg/util/client"
-	"github.com/openkruise/kruise/pkg/util/expectations"
-	utilfeature "github.com/openkruise/kruise/pkg/util/feature"
-	"github.com/openkruise/kruise/pkg/util/requeueduration"
-	"github.com/openkruise/kruise/pkg/util/revision"
 	apps "k8s.io/api/apps/v1"
 	v1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -37,6 +30,14 @@ import (
 	kubecontroller "k8s.io/kubernetes/pkg/controller"
 	"k8s.io/utils/integer"
 	"sigs.k8s.io/controller-runtime/pkg/client"
+
+	appsv1alpha1 "github.com/openkruise/kruise/apis/apps/v1alpha1"
+	"github.com/openkruise/kruise/pkg/features"
+	utilclient "github.com/openkruise/kruise/pkg/util/client"
+	"github.com/openkruise/kruise/pkg/util/expectations"
+	utilfeature "github.com/openkruise/kruise/pkg/util/feature"
+	"github.com/openkruise/kruise/pkg/util/requeueduration"
+	"github.com/openkruise/kruise/pkg/util/revision"
 )
 
 var (

pkg/controller/statefulset/stateful_set_control.go

@@ -45,7 +45,6 @@ import (
 	imagejobutilfunc "github.com/openkruise/kruise/pkg/util/imagejob/utilfunction"
 	"github.com/openkruise/kruise/pkg/util/inplaceupdate"
 	"github.com/openkruise/kruise/pkg/util/lifecycle"
-	"github.com/openkruise/kruise/pkg/util/pvc"
 )
 
 // Realistic value for maximum in-flight requests when processing in parallel mode.
@@ -610,12 +609,8 @@ func (ssc *defaultStatefulSetControl) rollingUpdateStatefulsetPods(
 			}
 		} else if utilfeature.DefaultFeatureGate.Enabled(features.StatefulSetAutoResizePVCGate) &&
 			set.Spec.VolumeClaimUpdateStrategy.Type == appsv1beta1.OnPodRollingUpdateVolumeClaimUpdateStrategyType {
-			checkReadyFn := func(claim, template *v1.PersistentVolumeClaim) bool {
-				_, ready := pvc.IsPVCCompatibleAndReady(claim, template)
-				return ready
-			}
 			// check pvc resize status, if not ready, record pod to unavailablePods
-			ready, err := ssc.checkOwnedPVCStatus(set, replicas[target], checkReadyFn)
+			ready, err := ssc.podControl.IsOwnedPVCsReady(set, replicas[target])
 			if err == nil && ready {
 				continue
 			}
@@ -645,30 +640,34 @@ func (ssc *defaultStatefulSetControl) rollingUpdateStatefulsetPods(
 			return status, nil
 		}
 
-		// delete the Pod if it is not already terminating and does not match the update revision.
-		if !isTerminating(replicas[target]) {
-			if utilfeature.DefaultFeatureGate.Enabled(features.StatefulSetAutoResizePVCGate) &&
-				set.Spec.VolumeClaimUpdateStrategy.Type == appsv1beta1.OnPodRollingUpdateVolumeClaimUpdateStrategyType {
-				// resize pvc if necessary and wait for resize completed
-				if match, err := ssc.podControl.IsClaimsCompatible(set, replicas[target]); err != nil {
-					return status, err
-				} else if !match {
-					err = ssc.podControl.TryPatchPVCSize(set, replicas[target])
-					if err != nil {
-						return status, err
-					}
-				}
-
-				allCompleted, err := ssc.checkOwnedPVCStatus(set, replicas[target], pvc.IsPatchPVCCompleted)
+		// Kruise currently will not patch pvc size until a pod references the resized volume.
+		// online-file-system-expansion: if no pods referencing the volume are running, file system expansion will not happen.
+		// refer to https://kubernetes.io/blog/2018/07/12/resizing-persistent-volumes-using-kubernetes/#online-file-system-expansion
+		if utilfeature.DefaultFeatureGate.Enabled(features.StatefulSetAutoResizePVCGate) &&
+			set.Spec.VolumeClaimUpdateStrategy.Type == appsv1beta1.OnPodRollingUpdateVolumeClaimUpdateStrategyType {
+			// resize pvc if necessary and wait for resize completed
+			if match, err := ssc.podControl.IsClaimsCompatible(set, replicas[target]); err != nil {
+				return status, err
+			} else if !match {
+				err = ssc.podControl.TryPatchPVC(set, replicas[target])
 				if err != nil {
 					return status, err
-				} else if !allCompleted {
-					// mark target as unavailable because pvc's updated
-					unavailablePods.Insert(replicas[target].Name)
-					// need to wait for pvc resize completed, continue to handle next pod
-					continue
 				}
 			}
+
+			allCompleted, err := ssc.podControl.IsOwnedPVCsCompleted(set, replicas[target])
+			if err != nil {
+				return status, err
+			} else if !allCompleted {
+				// mark target as unavailable because pvc's updated
+				unavailablePods.Insert(replicas[target].Name)
+				// need to wait for pvc resize completed, continue to handle next pod
+				continue
+			}
+		}
+
+		// delete the Pod if it is not already terminating and does not match the update revision.
+		if !isTerminating(replicas[target]) {
 			// todo validate in-place for pub
 			inplacing, inplaceUpdateErr := ssc.inPlaceUpdatePod(set, replicas[target], updateRevision, revisions)
 			if inplaceUpdateErr != nil {

pkg/controller/statefulset/stateful_set_utils.go

@@ -104,7 +104,7 @@ func getPodName(set *appsv1beta1.StatefulSet, ordinal int) string {
 }
 
 // getPersistentVolumeClaimName gets the name of PersistentVolumeClaim for a Pod with an ordinal index of ordinal. claim
-// must be a PersistentVolumeClaim from set's VolumeClaims template.
+// must be a PersistentVolumeClaim from set's VolumeClaimTemplates template.
 func getPersistentVolumeClaimName(set *appsv1beta1.StatefulSet, claim *v1.PersistentVolumeClaim, ordinal int) string {
 	// NOTE: This name format is used by the heuristics for zone spreading in ChooseZoneForVolume
 	return fmt.Sprintf("%s-%s-%d", claim.Name, set.Name, ordinal)
@@ -623,15 +623,32 @@ func nextRevision(revisions []*apps.ControllerRevision) int64 {
 // inconsistentStatus returns true if the ObservedGeneration of status is greater than set's
 // Generation or if any of the status's fields do not match those of set's status.
 func inconsistentStatus(set *appsv1beta1.StatefulSet, status *appsv1beta1.StatefulSetStatus) bool {
-	return status.ObservedGeneration > set.Status.ObservedGeneration ||
+	if status.ObservedGeneration > set.Status.ObservedGeneration ||
 		status.Replicas != set.Status.Replicas ||
 		status.CurrentReplicas != set.Status.CurrentReplicas ||
 		status.ReadyReplicas != set.Status.ReadyReplicas ||
 		status.AvailableReplicas != set.Status.AvailableReplicas ||
 		status.UpdatedReplicas != set.Status.UpdatedReplicas ||
 		status.CurrentRevision != set.Status.CurrentRevision ||
 		status.UpdateRevision != set.Status.UpdateRevision ||
-		status.LabelSelector != set.Status.LabelSelector
+		status.LabelSelector != set.Status.LabelSelector {
+		return true
+	}
+
+	volumeClaimName2StatusIdx := map[string]int{}
+	for i, v := range status.VolumeClaims {
+		volumeClaimName2StatusIdx[v.VolumeClaimName] = i
+	}
+	for _, v := range set.Status.VolumeClaims {
+		if idx, ok := volumeClaimName2StatusIdx[v.VolumeClaimName]; !ok {
+			// raw template not exist in current status => inconsistent
+			return true
+		} else if status.VolumeClaims[idx].CompatibleReplicas != v.CompatibleReplicas ||
+			status.VolumeClaims[idx].CompatibleReadyReplicas != v.CompatibleReadyReplicas {
+			return true
+		}
+	}
+	return false
 }
 
 // completeRollingUpdate completes a rolling update when all of set's replica Pods have been updated