fix: Fix dynamic release partition may fail search/query request

cause concurrent issue may occur between remove parition in target manager and sync segment list to delegator. when it happens, some segment may be released in delegator, and those segment may also be synced to delegator, which cause delegator become unserviceable due to lack of necessary segments, then search/query fails. this PR make sure that all write access to target_manager will be executed in serial to avoid the concurrent issues. Signed-off-by: Wei Liu <[email protected]>
milvus-io · Sep 3, 2024 · 7ee41a8 · 7ee41a8
1 parent 9d80137
commit 7ee41a8
Show file tree

Hide file tree

Showing 5 changed files with 95 additions and 30 deletions.
diff --git a/internal/querycoordv2/job/job_release.go b/internal/querycoordv2/job/job_release.go
@@ -95,7 +95,6 @@ func (job *ReleaseCollectionJob) Execute() error {
 		log.Warn(msg, zap.Error(err))
 	}
 
-	job.targetMgr.RemoveCollection(req.GetCollectionID())
 	job.targetObserver.ReleaseCollection(req.GetCollectionID())
 	waitCollectionReleased(job.dist, job.checkerController, req.GetCollectionID())
 	metrics.QueryCoordNumCollections.WithLabelValues().Dec()
@@ -178,7 +177,6 @@ func (job *ReleasePartitionJob) Execute() error {
 		if err != nil {
 			log.Warn("failed to remove replicas", zap.Error(err))
 		}
-		job.targetMgr.RemoveCollection(req.GetCollectionID())
 		job.targetObserver.ReleaseCollection(req.GetCollectionID())
 		metrics.QueryCoordNumCollections.WithLabelValues().Dec()
 		waitCollectionReleased(job.dist, job.checkerController, req.GetCollectionID())
@@ -189,7 +187,7 @@ func (job *ReleasePartitionJob) Execute() error {
 			log.Warn(msg, zap.Error(err))
 			return errors.Wrap(err, msg)
 		}
-		job.targetMgr.RemovePartition(req.GetCollectionID(), toRelease...)
+		job.targetObserver.ReleasePartition(req.GetCollectionID(), toRelease...)
 		waitCollectionReleased(job.dist, job.checkerController, req.GetCollectionID(), toRelease...)
 	}
 	metrics.QueryCoordNumPartitions.WithLabelValues().Sub(float64(len(toRelease)))

diff --git a/internal/querycoordv2/job/undo.go b/internal/querycoordv2/job/undo.go
@@ -78,10 +78,9 @@ func (u *UndoList) RollBack() {
 
 	if u.IsTargetUpdated {
 		if u.IsNewCollection {
-			u.targetMgr.RemoveCollection(u.CollectionID)
 			u.targetObserver.ReleaseCollection(u.CollectionID)
 		} else {
-			u.targetMgr.RemovePartition(u.CollectionID, u.LackPartitions...)
+			u.targetObserver.ReleasePartition(u.CollectionID, u.LackPartitions...)
 		}
 	}
 }
diff --git a/internal/querycoordv2/observers/collection_observer.go b/internal/querycoordv2/observers/collection_observer.go
@@ -180,7 +180,7 @@ func (ob *CollectionObserver) observeTimeout() {
 					zap.Duration("loadTime", time.Since(collection.CreatedAt)))
 				ob.meta.CollectionManager.RemoveCollection(collection.GetCollectionID())
 				ob.meta.ReplicaManager.RemoveCollection(collection.GetCollectionID())
-				ob.targetMgr.RemoveCollection(collection.GetCollectionID())
+				ob.targetObserver.ReleaseCollection(collection.GetCollectionID())
 				ob.loadTasks.Remove(traceID)
 			}
 		case querypb.LoadType_LoadPartition:
@@ -214,7 +214,7 @@ func (ob *CollectionObserver) observeTimeout() {
 					zap.Int64s("partitionIDs", task.PartitionIDs))
 				for _, partition := range partitions {
 					ob.meta.CollectionManager.RemovePartition(partition.CollectionID, partition.GetPartitionID())
-					ob.targetMgr.RemovePartition(partition.GetCollectionID(), partition.GetPartitionID())
+					ob.targetObserver.ReleasePartition(partition.GetCollectionID(), partition.GetPartitionID())
 				}
 
 				// all partition timeout, remove collection
@@ -223,7 +223,7 @@ func (ob *CollectionObserver) observeTimeout() {
 
 					ob.meta.CollectionManager.RemoveCollection(task.CollectionID)
 					ob.meta.ReplicaManager.RemoveCollection(task.CollectionID)
-					ob.targetMgr.RemoveCollection(task.CollectionID)
+					ob.targetObserver.ReleaseCollection(task.CollectionID)
 				}
 			}
 		}

diff --git a/internal/querycoordv2/observers/target_observer.go b/internal/querycoordv2/observers/target_observer.go
@@ -38,15 +38,33 @@ import (
 	"github.com/milvus-io/milvus/pkg/util/typeutil"
 )
 
-type checkRequest struct {
-	CollectionID int64
-	Notifier     chan bool
+type targetOp int
+
+func (op *targetOp) String() string {
+	switch *op {
+	case UpdateCollection:
+		return "UpdateCollection"
+	case ReleaseCollection:
+		return "ReleaseCollection"
+	case ReleasePartition:
+		return "ReleasePartition"
+	default:
+		return "Unknown"
+	}
 }
 
+const (
+	UpdateCollection targetOp = iota + 1
+	ReleaseCollection
+	ReleasePartition
+)
+
 type targetUpdateRequest struct {
 	CollectionID  int64
+	PartitionIDs  []int64
 	Notifier      chan error
 	ReadyNotifier chan struct{}
+	opType        targetOp
 }
 
 type initRequest struct{}
@@ -60,8 +78,7 @@ type TargetObserver struct {
 	broker    meta.Broker
 	cluster   session.Cluster
 
-	initChan    chan initRequest
-	manualCheck chan checkRequest
+	initChan chan initRequest
 	// nextTargetLastUpdate map[int64]time.Time
 	nextTargetLastUpdate *typeutil.ConcurrentMap[int64, time.Time]
 	updateChan           chan targetUpdateRequest
@@ -88,7 +105,6 @@ func NewTargetObserver(
 		distMgr:              distMgr,
 		broker:               broker,
 		cluster:              cluster,
-		manualCheck:          make(chan checkRequest, 10),
 		nextTargetLastUpdate: typeutil.NewConcurrentMap[int64, time.Time](),
 		updateChan:           make(chan targetUpdateRequest),
 		readyNotifiers:       make(map[int64][]chan struct{}),
@@ -152,23 +168,37 @@ func (ob *TargetObserver) schedule(ctx context.Context) {
 			ob.dispatcher.AddTask(ob.meta.GetAll()...)
 
 		case req := <-ob.updateChan:
-			log := log.With(zap.Int64("collectionID", req.CollectionID))
-			log.Info("manually trigger update next target")
-			ob.keylocks.Lock(req.CollectionID)
-			err := ob.updateNextTarget(req.CollectionID)
-			ob.keylocks.Unlock(req.CollectionID)
-			if err != nil {
-				log.Warn("failed to manually update next target", zap.Error(err))
-				close(req.ReadyNotifier)
-			} else {
-				ob.mut.Lock()
-				ob.readyNotifiers[req.CollectionID] = append(ob.readyNotifiers[req.CollectionID], req.ReadyNotifier)
-				ob.mut.Unlock()
+			log.Info("manually trigger update target",
+				zap.Int64("collectionID", req.CollectionID),
+				zap.String("opType", req.opType.String()),
+			)
+			switch req.opType {
+			case UpdateCollection:
+				ob.keylocks.Lock(req.CollectionID)
+				err := ob.updateNextTarget(req.CollectionID)
+				ob.keylocks.Unlock(req.CollectionID)
+				if err != nil {
+					log.Warn("failed to manually update next target",
+						zap.Int64("collectionID", req.CollectionID),
+						zap.String("opType", req.opType.String()),
+						zap.Error(err))
+					close(req.ReadyNotifier)
+				} else {
+					ob.mut.Lock()
+					ob.readyNotifiers[req.CollectionID] = append(ob.readyNotifiers[req.CollectionID], req.ReadyNotifier)
+					ob.mut.Unlock()
+				}
+				req.Notifier <- err
+			case ReleaseCollection:
+				ob.targetMgr.RemoveCollection(req.CollectionID)
+				req.Notifier <- nil
+			case ReleasePartition:
+				ob.targetMgr.RemovePartition(req.CollectionID, req.PartitionIDs...)
+				req.Notifier <- nil
 			}
-
-			log.Info("manually trigger update target done")
-			req.Notifier <- err
-			log.Info("notify manually trigger update target done")
+			log.Info("manually trigger update target done",
+				zap.Int64("collectionID", req.CollectionID),
+				zap.String("opType", req.opType.String()))
 		}
 	}
 }
@@ -229,6 +259,7 @@ func (ob *TargetObserver) UpdateNextTarget(collectionID int64) (chan struct{}, e
 
 	ob.updateChan <- targetUpdateRequest{
 		CollectionID:  collectionID,
+		opType:        UpdateCollection,
 		Notifier:      notifier,
 		ReadyNotifier: readyCh,
 	}
@@ -242,6 +273,28 @@ func (ob *TargetObserver) ReleaseCollection(collectionID int64) {
 		close(notifier)
 	}
 	delete(ob.readyNotifiers, collectionID)
+
+	notifier := make(chan error)
+	ob.updateChan <- targetUpdateRequest{
+		CollectionID: collectionID,
+		opType:       ReleaseCollection,
+		Notifier:     notifier,
+	}
+	<-notifier
+}
+
+func (ob *TargetObserver) ReleasePartition(collectionID int64, partitionID ...int64) {
+	ob.mut.Lock()
+	defer ob.mut.Unlock()
+
+	notifier := make(chan error)
+	ob.updateChan <- targetUpdateRequest{
+		CollectionID: collectionID,
+		PartitionIDs: partitionID,
+		opType:       ReleasePartition,
+		Notifier:     notifier,
+	}
+	<-notifier
 }
 
 func (ob *TargetObserver) clean() {

diff --git a/internal/querycoordv2/observers/target_observer_test.go b/internal/querycoordv2/observers/target_observer_test.go
@@ -21,6 +21,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/samber/lo"
 	"github.com/stretchr/testify/mock"
 	"github.com/stretchr/testify/suite"
 
@@ -214,6 +215,20 @@ func (suite *TargetObserverSuite) TestTriggerUpdateTarget() {
 	}, 7*time.Second, 1*time.Second)
 }
 
+func (suite *TargetObserverSuite) TestTriggerRelease() {
+	// Manually update next target
+	_, err := suite.observer.UpdateNextTarget(suite.collectionID)
+	suite.NoError(err)
+
+	// manually release partition
+	partitions := suite.meta.CollectionManager.GetPartitionsByCollection(suite.collectionID)
+	partitionIDs := lo.Map(partitions, func(partition *meta.Partition, _ int) int64 { return partition.PartitionID })
+	suite.observer.ReleasePartition(suite.collectionID, partitionIDs[0])
+
+	// manually release collection
+	suite.observer.ReleaseCollection(suite.collectionID)
+}
+
 func (suite *TargetObserverSuite) TearDownTest() {
 	suite.kv.Close()
 	suite.observer.Stop()