Longhorn Online Rebuilding

controller/engine_controller.go

@@ -1948,6 +1948,12 @@
 				"Start rebuilding replica %v with Address %v for normal engine %v and volume %v", replicaName, addr, e.Name, e.Spec.VolumeName)
 			err = engineClientProxy.ReplicaAdd(e, replicaName, replicaURL, false, fastReplicaRebuild, localSync, fileSyncHTTPClientTimeout, grpcTimeoutSeconds)
 		}
+
+		// For v2 engine, the rebuilding is an async call. We need to wait for the rebuilding start then complete here
+		if err == nil && types.IsDataEngineV2(e.Spec.DataEngine) {
+			err = ec.waitForV2EngineRebuild(e, replicaName, grpcTimeoutSeconds)
+		}
+
 		if err != nil {
 			replicaRebuildErrMsg := err.Error()
 
@@ -2121,6 +2127,64 @@
 	}
 }
 
+func (ec *EngineController) waitForV2EngineRebuild(e *longhorn.Engine, replicaName string, timeout int64) (err error) {
+	if !types.IsDataEngineV2(e.Spec.DataEngine) {
+		return nil
+	}
+
+	ticker := time.NewTicker(EnginePollInterval)
+	defer ticker.Stop()
+	timer := time.NewTimer(time.Duration(timeout) * time.Second)
+	defer timer.Stop()
+	for {
+		select {
+		case <-ticker.C:
+			e, err = ec.ds.GetEngineRO(e.Name)
+			if err != nil {
+				// There is no need to continue if the engine is not found
+				if apierrors.IsNotFound(err) {
+					return err
+				}
+				// There may be something wrong with the indexer or the API sever, will retry
+				continue
+			}
+			if e.Spec.ReplicaAddressMap[replicaName] == "" {
+				return fmt.Errorf("unknown replica %v for engine", replicaName)
+			}
+			// There is no need to continue when the replica is not found or the replica is not in a valid state for rebuilding
+			r, err := ec.ds.GetReplicaRO(replicaName)
+			if err != nil {
+				return err
+			}
+			if r.Status.CurrentState != longhorn.InstanceStateRunning {
+				return fmt.Errorf("replica %v is state %s, which is invalid for rebuilding", replicaName, r.Status.CurrentState)
+			}
+			if e.Status.ReplicaModeMap[replicaName] == longhorn.ReplicaModeRW {
+				return nil
+			}
+			if e.Status.ReplicaModeMap[replicaName] == longhorn.ReplicaModeERR {
+				return fmt.Errorf("replica %v is in ERR mode, which is invalid for rebuilding", replicaName)
+			}
+			if e.Status.ReplicaModeMap[replicaName] == "" {
+				continue
+			}
+			// For a rebuilding replica (with mode WO), there should be a corresponding rebuilding status
+			rebuildingStatus := e.Status.RebuildStatus[engineapi.GetBackendReplicaURL(e.Status.CurrentReplicaAddressMap[replicaName])]
+			if rebuildingStatus == nil {
+				continue
+			}
+			if rebuildingStatus.State == engineapi.ProcessStateError || rebuildingStatus.Error != "" {
+				return fmt.Errorf(rebuildingStatus.Error)
+			}
+			if rebuildingStatus.State == engineapi.ProcessStateComplete {
+				return nil
+			}
+		case <-timer.C:
+			return fmt.Errorf("timeout waiting for replica %v to be rebuilt", replicaName)
+		}
+	}
+}
+
 func (ec *EngineController) Upgrade(e *longhorn.Engine, log *logrus.Entry) (err error) {
 	defer func() {
 		err = errors.Wrapf(err, "failed to live upgrade image for %v", e.Name)

controller/replica_controller.go

@@ -386,7 +386,6 @@ func (rc *ReplicaController) CreateInstance(obj interface{}) (*longhorn.Instance
 		DataPath:            dataPath,
 		BackingImagePath:    backingImagePath,
 		DataLocality:        v.Spec.DataLocality,
-		ExposeRequired:      true,
 		ImIP:                im.Status.IP,
 		EngineCLIAPIVersion: cliAPIVersion,
 	})

controller/volume_controller.go

@@ -3640,12 +3640,6 @@ func (c *VolumeController) createReplica(replica *longhorn.Replica, v *longhorn.
 
 	if isRebuildingReplica {
 		// TODO: reuse failed replica for replica rebuilding of SPDK volumes
-		if types.IsDataEngineV2(v.Spec.DataEngine) {
-			if !v.Spec.DisableFrontend {
-				log.Tracef("Online replica rebuilding for replica %v is not supported for SPDK volumes", replica.Name)
-				return nil
-			}
-		}
 
 		log.Infof("A new replica %v will be replenished during rebuilding", replica.Name)
 		// Prevent this new replica from being reused after rebuilding failure.
@@ -4842,6 +4836,10 @@ func (c *VolumeController) shouldCleanUpFailedReplica(v *longhorn.Volume, r *lon
 		log.Warnf("Replica %v failed to rebuild too many times", r.Name)
 		return true
 	}
+	// TODO: Remove it once we can reuse failed replicas during v2 rebuilding
+	if types.IsDataEngineV2(v.Spec.DataEngine) {
+		return true
+	}
 	// Failed too long ago to be useful during a rebuild.
 	if v.Spec.StaleReplicaTimeout > 0 &&
 		util.TimestampAfterTimeout(r.Spec.FailedAt, time.Duration(v.Spec.StaleReplicaTimeout)*time.Minute) {

engineapi/instance_manager.go

@@ -517,7 +517,6 @@ type ReplicaInstanceCreateRequest struct {
 	DataPath            string
 	BackingImagePath    string
 	DataLocality        longhorn.DataLocality
-	ExposeRequired      bool
 	ImIP                string
 	EngineCLIAPIVersion int
 }
@@ -562,9 +561,8 @@ func (c *InstanceManagerClient) ReplicaInstanceCreate(req *ReplicaInstanceCreate
 		BinaryArgs: args,
 
 		Replica: imclient.ReplicaCreateRequest{
-			DiskName:       req.DiskName,
-			DiskUUID:       req.Replica.Spec.DiskID,
-			ExposeRequired: req.ExposeRequired,
+			DiskName: req.DiskName,
+			DiskUUID: req.Replica.Spec.DiskID,
 		},
 	})
 	if err != nil {

go.mod

@@ -62,9 +62,9 @@ require (
 	github.com/longhorn/backupstore v0.0.0-20240720163059-56c90cd23634
 	github.com/longhorn/go-common-libs v0.0.0-20240720044518-32fc527fe868
 	github.com/longhorn/go-iscsi-helper v0.0.0-20240720064937-c6ce82d67032
-	github.com/longhorn/go-spdk-helper v0.0.0-20240720064915-d2ce0846d2a7
+	github.com/longhorn/go-spdk-helper v0.0.0-20240723045100-85e5737da4b5
 	github.com/longhorn/longhorn-engine v1.7.0-rc1
-	github.com/longhorn/longhorn-instance-manager v1.7.0-rc1
+	github.com/longhorn/longhorn-instance-manager v1.7.0-dev.0.20240724045856-8b48e7b74784
 	github.com/longhorn/longhorn-share-manager v1.7.0-rc1
 	github.com/pkg/errors v0.9.1
 	github.com/prometheus/client_golang v1.18.0
@@ -103,7 +103,7 @@ require (
 	github.com/google/gnostic-models v0.6.8 // indirect
 	github.com/google/pprof v0.0.0-20231023181126-ff6d637d2a7b // indirect
 	github.com/jonboulle/clockwork v0.4.0 // indirect
-	github.com/longhorn/types v0.0.0-20240706151541-33cb010c3544 // indirect
+	github.com/longhorn/types v0.0.0-20240723142222-56701c990023 // indirect
 	github.com/mitchellh/go-ps v1.0.0 // indirect
 	github.com/moby/term v0.0.0-20221205130635-1aeaba878587 // indirect
 	github.com/power-devops/perfstat v0.0.0-20240221224432-82ca36839d55 // indirect

go.sum

@@ -1228,16 +1228,16 @@ github.com/longhorn/go-common-libs v0.0.0-20240720044518-32fc527fe868 h1:B1L6EGH
 github.com/longhorn/go-common-libs v0.0.0-20240720044518-32fc527fe868/go.mod h1:o01gaAiKE5NCd8+5i6csJSU4ELlW0Yn8GQ9U7pbXG9w=
 github.com/longhorn/go-iscsi-helper v0.0.0-20240720064937-c6ce82d67032 h1:93y5fe2y/TeP58k6VDsVrGHTy+dbaJ28DaNRGhrnYs8=
 github.com/longhorn/go-iscsi-helper v0.0.0-20240720064937-c6ce82d67032/go.mod h1:r72HLu6d9YRW5u5ovmw8IkUFQoy9zHObx8yn9DCFp/w=
-github.com/longhorn/go-spdk-helper v0.0.0-20240720064915-d2ce0846d2a7 h1:bwieBy2bUVi6sxFi2UbpXt292RykALG31yuChEYo66w=
-github.com/longhorn/go-spdk-helper v0.0.0-20240720064915-d2ce0846d2a7/go.mod h1:BxsYHvCYn/1rfohn5021Wt8tFYQb+unq1jItx8+htHY=
+github.com/longhorn/go-spdk-helper v0.0.0-20240723045100-85e5737da4b5 h1:zz0gXSR3pjZ+iBPym+rfpsu+tMGd5TfgOsgvue867cc=
+github.com/longhorn/go-spdk-helper v0.0.0-20240723045100-85e5737da4b5/go.mod h1:lt/eCeOltUM9QbMxS6qDJRr/+wvavxbLGTzkKrqyaSI=
 github.com/longhorn/longhorn-engine v1.7.0-rc1 h1:kQ4BwCvBemWsQvVnLdRsbiCXR3zmQBMA1J9Ezysh2q0=
 github.com/longhorn/longhorn-engine v1.7.0-rc1/go.mod h1:u0TZ1221YusDYA+ExdVLjLid1Ps6JuJXgh9185l5D9Y=
-github.com/longhorn/longhorn-instance-manager v1.7.0-rc1 h1:9hugpEQEmK6tMa1zY87GX1xrdq1Wb3TDqNl5LQzCWhM=
-github.com/longhorn/longhorn-instance-manager v1.7.0-rc1/go.mod h1:HIo3UCiH81EtTPwOujkKZIIQbvmJ1A3OeRHShHlAEV0=
+github.com/longhorn/longhorn-instance-manager v1.7.0-dev.0.20240724045856-8b48e7b74784 h1:LVtcsdIlxWcISUN5uMG9C4x9qQlRfDTNbaLFL3+R5H0=
+github.com/longhorn/longhorn-instance-manager v1.7.0-dev.0.20240724045856-8b48e7b74784/go.mod h1:gLVS19NOiRcfxJoGBTVVdY6TntbXi/Ks6nWofoJZQsM=
 github.com/longhorn/longhorn-share-manager v1.7.0-rc1 h1:LsSkSajhG8tCfORKKfwK+8XHVrT/8rI9DRWb7fuoVls=
 github.com/longhorn/longhorn-share-manager v1.7.0-rc1/go.mod h1:R6+NscPU4lAV5ueO7//lBCAO3en0aDbZi5KkkOSUJvk=
-github.com/longhorn/types v0.0.0-20240706151541-33cb010c3544 h1:U08l+0SbxCsododsraBHB5PdXrQme3TEh9iaREhRLQs=
-github.com/longhorn/types v0.0.0-20240706151541-33cb010c3544/go.mod h1:KlJuZB8NfHchWshYxYgV9pPIxBKC04Vq05G2TfgMf7w=
+github.com/longhorn/types v0.0.0-20240723142222-56701c990023 h1:6BYl+2rZhXjYZiBeLnWZPUgOiEoXGFA9CjcIfETdP/o=
+github.com/longhorn/types v0.0.0-20240723142222-56701c990023/go.mod h1:KlJuZB8NfHchWshYxYgV9pPIxBKC04Vq05G2TfgMf7w=
 github.com/lyft/protoc-gen-star v0.6.0/go.mod h1:TGAoBVkt8w7MPG72TrKIu85MIdXwDuzJYeZuUPFPNwA=
 github.com/lyft/protoc-gen-star v0.6.1/go.mod h1:TGAoBVkt8w7MPG72TrKIu85MIdXwDuzJYeZuUPFPNwA=
 github.com/lyft/protoc-gen-star/v2 v2.0.1/go.mod h1:RcCdONR2ScXaYnQC5tUzxzlpA3WVYF7/opLeUgcQs/o=

scheduler/replica_scheduler.go

@@ -566,6 +566,11 @@ func filterActiveReplicas(replicas map[string]*longhorn.Replica) map[string]*lon
 }
 
 func (rcs *ReplicaScheduler) CheckAndReuseFailedReplica(replicas map[string]*longhorn.Replica, volume *longhorn.Volume, hardNodeAffinity string) (*longhorn.Replica, error) {
+	// TODO: Remove it once we can reuse failed replicas during v2 rebuilding
+	if types.IsDataEngineV2(volume.Spec.DataEngine) {
+		return nil, nil
+	}
+
 	replicas = filterActiveReplicas(replicas)
 
 	allNodesInfo, err := rcs.getNodeInfo()
@@ -654,6 +659,11 @@ func (rcs *ReplicaScheduler) RequireNewReplica(replicas map[string]*longhorn.Rep
 		return 0
 	}
 
+	// TODO: Remove it once we can reuse failed replicas during v2 rebuilding
+	if types.IsDataEngineV2(volume.Spec.DataEngine) {
+		return 0
+	}
+
 	timeUntilNext, timeOfNext, err := rcs.timeToReplacementReplica(volume)
 	if err != nil {
 		msg := "Failed to get time until replica replacement, will directly replenish a new replica"