Fix support for multiple pvc for pd (#3820)

docs/api-references/docs.md

@@ -7799,6 +7799,16 @@ k8s.io/apimachinery/pkg/types.UID
 </tr>
 <tr>
 <td>
+<code>pvcUIDSet</code></br>
+<em>
+map[k8s.io/apimachinery/pkg/types.UID]struct{}
+</em>
+</td>
+<td>
+</td>
+</tr>
+<tr>
+<td>
 <code>memberDeleted</code></br>
 <em>
 bool
@@ -9073,6 +9083,7 @@ map[string]github.com/pingcap/tidb-operator/pkg/apis/pingcap/v1alpha1.PDMember
 </em>
 </td>
 <td>
+<p>Members contains PDs in current TidbCluster</p>
 </td>
 </tr>
 <tr>
@@ -9085,6 +9096,7 @@ map[string]github.com/pingcap/tidb-operator/pkg/apis/pingcap/v1alpha1.PDMember
 </em>
 </td>
 <td>
+<p>PeerMembers contains PDs NOT in current TidbCluster</p>
 </td>
 </tr>
 <tr>

pkg/apis/pingcap/v1alpha1/tidbcluster.go

@@ -334,13 +334,13 @@ func (tc *TidbCluster) PDAutoFailovering() bool {
 }
 
 func (tc *TidbCluster) GetPDDeletedFailureReplicas() int32 {
-	var failureReplicas int32 = 0
+	var deteledReplicas int32 = 0
 	for _, failureMember := range tc.Status.PD.FailureMembers {
 		if failureMember.MemberDeleted {
-			failureReplicas++
+			deteledReplicas++
 		}
 	}
-	return failureReplicas
+	return deteledReplicas
 }
 
 func (tc *TidbCluster) PDStsDesiredReplicas() int32 {

pkg/apis/pingcap/v1alpha1/types.go

@@ -940,10 +940,12 @@ type Service struct {
 
 // PDStatus is PD status
 type PDStatus struct {
-	Synced          bool                       `json:"synced,omitempty"`
-	Phase           MemberPhase                `json:"phase,omitempty"`
-	StatefulSet     *apps.StatefulSetStatus    `json:"statefulSet,omitempty"`
-	Members         map[string]PDMember        `json:"members,omitempty"`
+	Synced      bool                    `json:"synced,omitempty"`
+	Phase       MemberPhase             `json:"phase,omitempty"`
+	StatefulSet *apps.StatefulSetStatus `json:"statefulSet,omitempty"`
+	// Members contains PDs in current TidbCluster
+	Members map[string]PDMember `json:"members,omitempty"`
+	// PeerMembers contains PDs NOT in current TidbCluster
 	PeerMembers     map[string]PDMember        `json:"peerMembers,omitempty"`
 	Leader          PDMember                   `json:"leader,omitempty"`
 	FailureMembers  map[string]PDFailureMember `json:"failureMembers,omitempty"`
@@ -965,11 +967,12 @@ type PDMember struct {
 
 // PDFailureMember is the pd failure member information
 type PDFailureMember struct {
-	PodName       string      `json:"podName,omitempty"`
-	MemberID      string      `json:"memberID,omitempty"`
-	PVCUID        types.UID   `json:"pvcUID,omitempty"`
-	MemberDeleted bool        `json:"memberDeleted,omitempty"`
-	CreatedAt     metav1.Time `json:"createdAt,omitempty"`
+	PodName       string                 `json:"podName,omitempty"`
+	MemberID      string                 `json:"memberID,omitempty"`
+	PVCUID        types.UID              `json:"pvcUID,omitempty"`
+	PVCUIDSet     map[types.UID]struct{} `json:"pvcUIDSet,omitempty"`
+	MemberDeleted bool                   `json:"memberDeleted,omitempty"`
+	CreatedAt     metav1.Time            `json:"createdAt,omitempty"`
 }
 
 // UnjoinedMember is the pd unjoin cluster member information

pkg/controller/tidbcluster/tidb_cluster_control.go

@@ -143,6 +143,7 @@ func (c *defaultTidbClusterControl) updateTidbCluster(tc *v1alpha1.TidbCluster)
 	}
 
 	// cleaning all orphan pods(pd, tikv or tiflash which don't have a related PVC) managed by operator
+	// this could be useful when failover run into an undesired situation as described in PD failover function
 	skipReasons, err := c.orphanPodsCleaner.Clean(tc)
 	if err != nil {
 		return err

pkg/manager/member/dm_master_member_manager.go

@@ -351,7 +351,7 @@ func (m *masterMemberManager) syncDMClusterStatus(dc *v1alpha1.DMCluster, set *a
 	dc.Status.Master.Members = masterStatus
 	dc.Status.Master.Leader = dc.Status.Master.Members[leader.Name]
 	dc.Status.Master.Image = ""
-	c := filterContainer(set, "dm-master")
+	c := findContainerByName(set, "dm-master")
 	if c != nil {
 		dc.Status.Master.Image = c.Image
 	}

pkg/manager/member/dm_master_scaler.go

@@ -61,7 +61,7 @@ func (s *masterScaler) ScaleOut(meta metav1.Object, oldSet *apps.StatefulSet, ne
 	dcName := dc.GetName()
 
 	klog.Infof("scaling out dm-master statefulset %s/%s, ordinal: %d (replicas: %d, delete slots: %v)", oldSet.Namespace, oldSet.Name, ordinal, replicas, deleteSlots.List())
-	_, err := s.deleteDeferDeletingPVC(dc, oldSet.GetName(), v1alpha1.DMMasterMemberType, ordinal)
+	_, err := s.deleteDeferDeletingPVC(dc, v1alpha1.DMMasterMemberType, ordinal)
 	if err != nil {
 		return err
 	}

pkg/manager/member/dm_worker_member_manager.go

@@ -280,7 +280,7 @@ func (m *workerMemberManager) syncDMClusterStatus(dc *v1alpha1.DMCluster, set *a
 	dc.Status.Worker.Synced = true
 	dc.Status.Worker.Members = workerStatus
 	dc.Status.Worker.Image = ""
-	c := filterContainer(set, "dm-worker")
+	c := findContainerByName(set, "dm-worker")
 	if c != nil {
 		dc.Status.Worker.Image = c.Image
 	}

pkg/manager/member/dm_worker_scaler.go

@@ -61,7 +61,7 @@ func (s *workerScaler) ScaleOut(meta metav1.Object, oldSet *apps.StatefulSet, ne
 	dcName := dc.GetName()
 
 	klog.Infof("scaling out dm-worker statefulset %s/%s, ordinal: %d (replicas: %d, delete slots: %v)", oldSet.Namespace, oldSet.Name, ordinal, replicas, deleteSlots.List())
-	_, err := s.deleteDeferDeletingPVC(dc, oldSet.GetName(), v1alpha1.DMWorkerMemberType, ordinal)
+	_, err := s.deleteDeferDeletingPVC(dc, v1alpha1.DMWorkerMemberType, ordinal)
 	if err != nil {
 		return err
 	}

pkg/manager/member/failover.go

@@ -15,6 +15,9 @@ package member
 
 import "github.com/pingcap/tidb-operator/pkg/apis/pingcap/v1alpha1"
 
+// TODO: move this to a centralized place
+// Since the "Unhealthy" is a very universal event reason string, which could apply to all the TiDB/DM cluster components,
+// we should make a global event module, and put event related constants there.
 const (
 	unHealthEventReason     = "Unhealthy"
 	unHealthEventMsgPattern = "%s pod[%s] is unhealthy, msg:%s"