From a447a1a37a6f42c9569d318edda45890ad8371e7 Mon Sep 17 00:00:00 2001
From: sunshuai09 <sunshuai09@meituan.com>
Date: Tue, 1 Nov 2022 19:47:25 +0800
Subject: [PATCH] enable mcloneset deleting pvc when pod hanging

---
 apis/apps/v1alpha1/cloneset_types.go          |  3 +
 .../crd/bases/apps.kruise.io_clonesets.yaml   |  4 ++
 .../apps.kruise.io_uniteddeployments.yaml     |  4 ++
 .../cloneset/sync/cloneset_scale.go           | 57 +++++++++++++++++++
 4 files changed, 68 insertions(+)

diff --git a/apis/apps/v1alpha1/cloneset_types.go b/apis/apps/v1alpha1/cloneset_types.go
index 4d95dd224e..82aee66999 100644
--- a/apis/apps/v1alpha1/cloneset_types.go
+++ b/apis/apps/v1alpha1/cloneset_types.go
@@ -93,6 +93,9 @@ type CloneSetScaleStrategy struct {
 	// The scale will fail if the number of unavailable pods were greater than this MaxUnavailable at scaling up.
 	// MaxUnavailable works only when scaling up.
 	MaxUnavailable *intstr.IntOrString `json:"maxUnavailable,omitempty"`
+
+	// Enable cloneset to clean up the useless pvc when pod hangs in terminating state.
+	CleanUpPVC bool `json:"cleanUpPVC,omitempty"`
 }
 
 // CloneSetUpdateStrategy defines strategies for pods update.
diff --git a/config/crd/bases/apps.kruise.io_clonesets.yaml b/config/crd/bases/apps.kruise.io_clonesets.yaml
index da2fb66cdc..d713f57ecd 100644
--- a/config/crd/bases/apps.kruise.io_clonesets.yaml
+++ b/config/crd/bases/apps.kruise.io_clonesets.yaml
@@ -149,6 +149,10 @@ spec:
                 description: ScaleStrategy indicates the ScaleStrategy that will be
                   employed to create and delete Pods in the CloneSet.
                 properties:
+                  cleanUpPVC:
+                    description: Enable cloneset to clean up the useless pvc when
+                      pod hangs in terminating state.
+                    type: boolean
                   maxUnavailable:
                     anyOf:
                     - type: integer
diff --git a/config/crd/bases/apps.kruise.io_uniteddeployments.yaml b/config/crd/bases/apps.kruise.io_uniteddeployments.yaml
index a274937b18..ea09182526 100644
--- a/config/crd/bases/apps.kruise.io_uniteddeployments.yaml
+++ b/config/crd/bases/apps.kruise.io_uniteddeployments.yaml
@@ -617,6 +617,10 @@ spec:
                               that will be employed to create and delete Pods in the
                               CloneSet.
                             properties:
+                              cleanUpPVC:
+                                description: Enable cloneset to clean up the useless
+                                  pvc when pod hangs in terminating state.
+                                type: boolean
                               maxUnavailable:
                                 anyOf:
                                 - type: integer
diff --git a/pkg/controller/cloneset/sync/cloneset_scale.go b/pkg/controller/cloneset/sync/cloneset_scale.go
index 69ca12c685..bfd2f73153 100644
--- a/pkg/controller/cloneset/sync/cloneset_scale.go
+++ b/pkg/controller/cloneset/sync/cloneset_scale.go
@@ -31,6 +31,7 @@ import (
 	"github.com/openkruise/kruise/pkg/util/expectations"
 	"github.com/openkruise/kruise/pkg/util/lifecycle"
 	v1 "k8s.io/api/core/v1"
+	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/apimachinery/pkg/util/rand"
 	"k8s.io/apimachinery/pkg/util/sets"
 	"k8s.io/klog/v2"
@@ -60,6 +61,21 @@ func (r *realControl) Scale(
 		return false, nil
 	}
 
+	// If cloneset doesn't want to reuse pvc, clean up the existing pvc first.
+	// Then it looks like the pod is deleted by controller, new pod with new name
+	// can be created.
+	if updateCS.Spec.ScaleStrategy.CleanUpPVC {
+		ins := getInstanceIDsFromPods(pods)
+		usingPVCs, uselessPVCs := classifyPVCs(ins, pvcs)
+		if len(uselessPVCs) > 0 {
+			klog.V(3).Infof("Begin to clean up cloneset %s useless PVCs", controllerKey)
+			if modified, err := r.cleanupPVCs(updateCS, uselessPVCs); err != nil || modified {
+				return modified, err
+			}
+			pvcs = usingPVCs
+		}
+	}
+
 	// 1. manage pods to delete and in preDelete
 	podsSpecifiedToDelete, podsInPreDelete, numToDelete := getPlannedDeletedPods(updateCS, pods)
 	if modified, err := r.managePreparingDelete(updateCS, pods, podsInPreDelete, numToDelete); err != nil || modified {
@@ -403,3 +419,44 @@ func (r *realControl) choosePodsToDelete(cs *appsv1alpha1.CloneSet, totalDiff in
 
 	return podsToDelete
 }
+
+func (r *realControl) cleanupPVCs(cs *appsv1alpha1.CloneSet, pvcs []*v1.PersistentVolumeClaim) (bool, error) {
+	var modified bool
+	for _, pvc := range pvcs {
+		clonesetutils.ScaleExpectations.ExpectScale(clonesetutils.GetControllerKey(cs), expectations.Delete, pvc.Name)
+		if err := r.Delete(context.TODO(), pvc); err != nil {
+			clonesetutils.ScaleExpectations.ObserveScale(clonesetutils.GetControllerKey(cs), expectations.Delete, pvc.Name)
+			r.recorder.Eventf(cs, v1.EventTypeWarning, "FailedCleanUp", "failed to clean up %s: %v", pvc.Name, err)
+			return modified, err
+		}
+	}
+	return modified, nil
+}
+
+func getInstanceIDsFromPods(pods []*v1.Pod) sets.String {
+	ins := sets.NewString()
+	for _, pod := range pods {
+		ins.Insert(pod.Labels[appsv1alpha1.CloneSetInstanceID])
+	}
+	return ins
+}
+
+func classifyPVCs(ids sets.String, pvcs []*v1.PersistentVolumeClaim) (using, useless []*v1.PersistentVolumeClaim) {
+	usingMap := map[types.UID]*v1.PersistentVolumeClaim{}
+	uselessMap := map[types.UID]*v1.PersistentVolumeClaim{}
+	for _, pvc := range pvcs {
+		if ids.Has(pvc.Labels[appsv1alpha1.CloneSetInstanceID]) {
+			usingMap[pvc.UID] = pvc
+		} else {
+			uselessMap[pvc.UID] = pvc
+		}
+	}
+
+	for _, p := range usingMap {
+		using = append(using, p)
+	}
+	for _, p := range uselessMap {
+		useless = append(useless, p)
+	}
+	return using, useless
+}