Fix ready notifiers leak when collection released (#21712)

Signed-off-by: yah01 <yang.cen@zilliz.com>
This commit is contained in:
yah01 2023-01-14 21:55:41 +08:00 committed by GitHub
parent aecb7ea639
commit 837e3162d7
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 100 additions and 36 deletions

View File

@ -28,6 +28,7 @@ import (
"github.com/milvus-io/milvus/internal/metrics" "github.com/milvus-io/milvus/internal/metrics"
"github.com/milvus-io/milvus/internal/proto/querypb" "github.com/milvus-io/milvus/internal/proto/querypb"
"github.com/milvus-io/milvus/internal/querycoordv2/meta" "github.com/milvus-io/milvus/internal/querycoordv2/meta"
"github.com/milvus-io/milvus/internal/querycoordv2/observers"
"github.com/milvus-io/milvus/internal/querycoordv2/session" "github.com/milvus-io/milvus/internal/querycoordv2/session"
"github.com/milvus-io/milvus/internal/querycoordv2/utils" "github.com/milvus-io/milvus/internal/querycoordv2/utils"
"github.com/milvus-io/milvus/internal/util/typeutil" "github.com/milvus-io/milvus/internal/util/typeutil"
@ -260,6 +261,7 @@ type ReleaseCollectionJob struct {
dist *meta.DistributionManager dist *meta.DistributionManager
meta *meta.Meta meta *meta.Meta
targetMgr *meta.TargetManager targetMgr *meta.TargetManager
targetObserver *observers.TargetObserver
} }
func NewReleaseCollectionJob(ctx context.Context, func NewReleaseCollectionJob(ctx context.Context,
@ -267,6 +269,7 @@ func NewReleaseCollectionJob(ctx context.Context,
dist *meta.DistributionManager, dist *meta.DistributionManager,
meta *meta.Meta, meta *meta.Meta,
targetMgr *meta.TargetManager, targetMgr *meta.TargetManager,
targetObserver *observers.TargetObserver,
) *ReleaseCollectionJob { ) *ReleaseCollectionJob {
return &ReleaseCollectionJob{ return &ReleaseCollectionJob{
BaseJob: NewBaseJob(ctx, req.Base.GetMsgID(), req.GetCollectionID()), BaseJob: NewBaseJob(ctx, req.Base.GetMsgID(), req.GetCollectionID()),
@ -274,6 +277,7 @@ func NewReleaseCollectionJob(ctx context.Context,
dist: dist, dist: dist,
meta: meta, meta: meta,
targetMgr: targetMgr, targetMgr: targetMgr,
targetObserver: targetObserver,
} }
} }
@ -301,6 +305,7 @@ func (job *ReleaseCollectionJob) Execute() error {
} }
job.targetMgr.RemoveCollection(req.GetCollectionID()) job.targetMgr.RemoveCollection(req.GetCollectionID())
job.targetObserver.ReleaseCollection(req.GetCollectionID())
waitCollectionReleased(job.dist, req.GetCollectionID()) waitCollectionReleased(job.dist, req.GetCollectionID())
metrics.QueryCoordNumCollections.WithLabelValues().Dec() metrics.QueryCoordNumCollections.WithLabelValues().Dec()
return nil return nil
@ -466,6 +471,7 @@ type ReleasePartitionJob struct {
dist *meta.DistributionManager dist *meta.DistributionManager
meta *meta.Meta meta *meta.Meta
targetMgr *meta.TargetManager targetMgr *meta.TargetManager
targetObserver *observers.TargetObserver
} }
func NewReleasePartitionJob(ctx context.Context, func NewReleasePartitionJob(ctx context.Context,
@ -473,6 +479,7 @@ func NewReleasePartitionJob(ctx context.Context,
dist *meta.DistributionManager, dist *meta.DistributionManager,
meta *meta.Meta, meta *meta.Meta,
targetMgr *meta.TargetManager, targetMgr *meta.TargetManager,
targetObserver *observers.TargetObserver,
) *ReleasePartitionJob { ) *ReleasePartitionJob {
return &ReleasePartitionJob{ return &ReleasePartitionJob{
BaseJob: NewBaseJob(ctx, req.Base.GetMsgID(), req.GetCollectionID()), BaseJob: NewBaseJob(ctx, req.Base.GetMsgID(), req.GetCollectionID()),
@ -480,6 +487,7 @@ func NewReleasePartitionJob(ctx context.Context,
dist: dist, dist: dist,
meta: meta, meta: meta,
targetMgr: targetMgr, targetMgr: targetMgr,
targetObserver: targetObserver,
} }
} }
@ -527,6 +535,7 @@ func (job *ReleasePartitionJob) Execute() error {
log.Warn("failed to remove replicas", zap.Error(err)) log.Warn("failed to remove replicas", zap.Error(err))
} }
job.targetMgr.RemoveCollection(req.GetCollectionID()) job.targetMgr.RemoveCollection(req.GetCollectionID())
job.targetObserver.ReleaseCollection(req.GetCollectionID())
waitCollectionReleased(job.dist, req.GetCollectionID()) waitCollectionReleased(job.dist, req.GetCollectionID())
} else { } else {
err := job.meta.CollectionManager.RemovePartition(toRelease...) err := job.meta.CollectionManager.RemovePartition(toRelease...)

View File

@ -29,6 +29,7 @@ import (
"github.com/milvus-io/milvus/internal/proto/datapb" "github.com/milvus-io/milvus/internal/proto/datapb"
"github.com/milvus-io/milvus/internal/proto/querypb" "github.com/milvus-io/milvus/internal/proto/querypb"
"github.com/milvus-io/milvus/internal/querycoordv2/meta" "github.com/milvus-io/milvus/internal/querycoordv2/meta"
"github.com/milvus-io/milvus/internal/querycoordv2/observers"
. "github.com/milvus-io/milvus/internal/querycoordv2/params" . "github.com/milvus-io/milvus/internal/querycoordv2/params"
"github.com/milvus-io/milvus/internal/querycoordv2/session" "github.com/milvus-io/milvus/internal/querycoordv2/session"
"github.com/milvus-io/milvus/internal/util/etcd" "github.com/milvus-io/milvus/internal/util/etcd"
@ -55,6 +56,7 @@ type JobSuite struct {
dist *meta.DistributionManager dist *meta.DistributionManager
meta *meta.Meta meta *meta.Meta
targetMgr *meta.TargetManager targetMgr *meta.TargetManager
targetObserver *observers.TargetObserver
broker *meta.MockBroker broker *meta.MockBroker
nodeMgr *session.NodeManager nodeMgr *session.NodeManager
@ -131,6 +133,11 @@ func (suite *JobSuite) SetupTest() {
suite.dist = meta.NewDistributionManager() suite.dist = meta.NewDistributionManager()
suite.meta = meta.NewMeta(RandomIncrementIDAllocator(), suite.store) suite.meta = meta.NewMeta(RandomIncrementIDAllocator(), suite.store)
suite.targetMgr = meta.NewTargetManager(suite.broker, suite.meta) suite.targetMgr = meta.NewTargetManager(suite.broker, suite.meta)
suite.targetObserver = observers.NewTargetObserver(suite.meta,
suite.targetMgr,
suite.dist,
suite.broker,
)
suite.nodeMgr = session.NewNodeManager() suite.nodeMgr = session.NewNodeManager()
suite.nodeMgr.Add(&session.NodeInfo{}) suite.nodeMgr.Add(&session.NodeInfo{})
suite.scheduler = NewScheduler() suite.scheduler = NewScheduler()
@ -583,6 +590,7 @@ func (suite *JobSuite) TestReleaseCollection() {
suite.dist, suite.dist,
suite.meta, suite.meta,
suite.targetMgr, suite.targetMgr,
suite.targetObserver,
) )
suite.scheduler.Add(job) suite.scheduler.Add(job)
err := job.Wait() err := job.Wait()
@ -601,6 +609,7 @@ func (suite *JobSuite) TestReleaseCollection() {
suite.dist, suite.dist,
suite.meta, suite.meta,
suite.targetMgr, suite.targetMgr,
suite.targetObserver,
) )
suite.scheduler.Add(job) suite.scheduler.Add(job)
err := job.Wait() err := job.Wait()
@ -626,6 +635,7 @@ func (suite *JobSuite) TestReleasePartition() {
suite.dist, suite.dist,
suite.meta, suite.meta,
suite.targetMgr, suite.targetMgr,
suite.targetObserver,
) )
suite.scheduler.Add(job) suite.scheduler.Add(job)
err := job.Wait() err := job.Wait()
@ -650,6 +660,7 @@ func (suite *JobSuite) TestReleasePartition() {
suite.dist, suite.dist,
suite.meta, suite.meta,
suite.targetMgr, suite.targetMgr,
suite.targetObserver,
) )
suite.scheduler.Add(job) suite.scheduler.Add(job)
err := job.Wait() err := job.Wait()
@ -676,6 +687,7 @@ func (suite *JobSuite) TestReleasePartition() {
suite.dist, suite.dist,
suite.meta, suite.meta,
suite.targetMgr, suite.targetMgr,
suite.targetObserver,
) )
suite.scheduler.Add(job) suite.scheduler.Add(job)
err := job.Wait() err := job.Wait()
@ -843,6 +855,7 @@ func (suite *JobSuite) releaseAll() {
suite.dist, suite.dist,
suite.meta, suite.meta,
suite.targetMgr, suite.targetMgr,
suite.targetObserver,
) )
suite.scheduler.Add(job) suite.scheduler.Add(job)
err := job.Wait() err := job.Wait()

View File

@ -46,6 +46,7 @@ type TargetObserver struct {
nextTargetLastUpdate map[int64]time.Time nextTargetLastUpdate map[int64]time.Time
updateChan chan targetUpdateRequest updateChan chan targetUpdateRequest
mut sync.Mutex // Guard readyNotifiers
readyNotifiers map[int64][]chan struct{} // CollectionID -> Notifiers readyNotifiers map[int64][]chan struct{} // CollectionID -> Notifiers
stopOnce sync.Once stopOnce sync.Once
@ -91,6 +92,7 @@ func (ob *TargetObserver) schedule(ctx context.Context) {
return return
case <-ticker.C: case <-ticker.C:
ob.clean()
ob.tryUpdateTarget() ob.tryUpdateTarget()
case request := <-ob.updateChan: case request := <-ob.updateChan:
@ -98,7 +100,9 @@ func (ob *TargetObserver) schedule(ctx context.Context) {
if err != nil { if err != nil {
close(request.ReadyNotifier) close(request.ReadyNotifier)
} else { } else {
ob.mut.Lock()
ob.readyNotifiers[request.CollectionID] = append(ob.readyNotifiers[request.CollectionID], request.ReadyNotifier) ob.readyNotifiers[request.CollectionID] = append(ob.readyNotifiers[request.CollectionID], request.ReadyNotifier)
ob.mut.Unlock()
} }
request.Notifier <- err request.Notifier <- err
@ -122,6 +126,15 @@ func (ob *TargetObserver) UpdateNextTarget(collectionID int64) (chan struct{}, e
return readyCh, <-notifier return readyCh, <-notifier
} }
func (ob *TargetObserver) ReleaseCollection(collectionID int64) {
ob.mut.Lock()
defer ob.mut.Unlock()
for _, notifier := range ob.readyNotifiers[collectionID] {
close(notifier)
}
delete(ob.readyNotifiers, collectionID)
}
func (ob *TargetObserver) tryUpdateTarget() { func (ob *TargetObserver) tryUpdateTarget() {
collections := ob.meta.GetAll() collections := ob.meta.GetAll()
for _, collectionID := range collections { for _, collectionID := range collections {
@ -144,6 +157,21 @@ func (ob *TargetObserver) tryUpdateTarget() {
} }
} }
func (ob *TargetObserver) clean() {
collections := typeutil.NewSet(ob.meta.GetAll()...)
ob.mut.Lock()
defer ob.mut.Unlock()
for collectionID, notifiers := range ob.readyNotifiers {
if !collections.Contain(collectionID) {
for i := range notifiers {
close(notifiers[i])
}
delete(ob.readyNotifiers, collectionID)
}
}
}
func (ob *TargetObserver) shouldUpdateNextTarget(collectionID int64) bool { func (ob *TargetObserver) shouldUpdateNextTarget(collectionID int64) bool {
return !ob.targetMgr.IsNextTargetExist(collectionID) || ob.isNextTargetExpired(collectionID) return !ob.targetMgr.IsNextTargetExist(collectionID) || ob.isNextTargetExpired(collectionID)
} }
@ -213,6 +241,9 @@ func (ob *TargetObserver) updateCurrentTarget(collectionID int64) {
log.Warn("observer trigger update current target", log.Warn("observer trigger update current target",
zap.Int64("collectionID", collectionID)) zap.Int64("collectionID", collectionID))
ob.targetMgr.UpdateCollectionCurrentTarget(collectionID) ob.targetMgr.UpdateCollectionCurrentTarget(collectionID)
ob.mut.Lock()
defer ob.mut.Unlock()
notifiers := ob.readyNotifiers[collectionID] notifiers := ob.readyNotifiers[collectionID]
for _, notifier := range notifiers { for _, notifier := range notifiers {
close(notifier) close(notifier)

View File

@ -254,6 +254,7 @@ func (s *Server) ReleaseCollection(ctx context.Context, req *querypb.ReleaseColl
s.dist, s.dist,
s.meta, s.meta,
s.targetMgr, s.targetMgr,
s.targetObserver,
) )
s.jobScheduler.Add(releaseJob) s.jobScheduler.Add(releaseJob)
err := releaseJob.Wait() err := releaseJob.Wait()
@ -339,6 +340,7 @@ func (s *Server) ReleasePartitions(ctx context.Context, req *querypb.ReleasePart
s.dist, s.dist,
s.meta, s.meta,
s.targetMgr, s.targetMgr,
s.targetObserver,
) )
s.jobScheduler.Add(releaseJob) s.jobScheduler.Add(releaseJob)
err := releaseJob.Wait() err := releaseJob.Wait()

View File

@ -33,6 +33,7 @@ import (
"github.com/milvus-io/milvus/internal/querycoordv2/balance" "github.com/milvus-io/milvus/internal/querycoordv2/balance"
"github.com/milvus-io/milvus/internal/querycoordv2/job" "github.com/milvus-io/milvus/internal/querycoordv2/job"
"github.com/milvus-io/milvus/internal/querycoordv2/meta" "github.com/milvus-io/milvus/internal/querycoordv2/meta"
"github.com/milvus-io/milvus/internal/querycoordv2/observers"
"github.com/milvus-io/milvus/internal/querycoordv2/params" "github.com/milvus-io/milvus/internal/querycoordv2/params"
"github.com/milvus-io/milvus/internal/querycoordv2/session" "github.com/milvus-io/milvus/internal/querycoordv2/session"
"github.com/milvus-io/milvus/internal/querycoordv2/task" "github.com/milvus-io/milvus/internal/querycoordv2/task"
@ -65,6 +66,7 @@ type ServiceSuite struct {
meta *meta.Meta meta *meta.Meta
targetMgr *meta.TargetManager targetMgr *meta.TargetManager
broker *meta.MockBroker broker *meta.MockBroker
targetObserver *observers.TargetObserver
cluster *session.MockCluster cluster *session.MockCluster
nodeMgr *session.NodeManager nodeMgr *session.NodeManager
jobScheduler *job.Scheduler jobScheduler *job.Scheduler
@ -127,6 +129,12 @@ func (suite *ServiceSuite) SetupTest() {
suite.meta = meta.NewMeta(params.RandomIncrementIDAllocator(), suite.store) suite.meta = meta.NewMeta(params.RandomIncrementIDAllocator(), suite.store)
suite.broker = meta.NewMockBroker(suite.T()) suite.broker = meta.NewMockBroker(suite.T())
suite.targetMgr = meta.NewTargetManager(suite.broker, suite.meta) suite.targetMgr = meta.NewTargetManager(suite.broker, suite.meta)
suite.targetObserver = observers.NewTargetObserver(
suite.meta,
suite.targetMgr,
suite.dist,
suite.broker,
)
suite.nodeMgr = session.NewNodeManager() suite.nodeMgr = session.NewNodeManager()
for _, node := range suite.nodes { for _, node := range suite.nodes {
suite.nodeMgr.Add(session.NewNodeInfo(node, "localhost")) suite.nodeMgr.Add(session.NewNodeInfo(node, "localhost"))
@ -153,6 +161,7 @@ func (suite *ServiceSuite) SetupTest() {
meta: suite.meta, meta: suite.meta,
targetMgr: suite.targetMgr, targetMgr: suite.targetMgr,
broker: suite.broker, broker: suite.broker,
targetObserver: suite.targetObserver,
nodeMgr: suite.nodeMgr, nodeMgr: suite.nodeMgr,
cluster: suite.cluster, cluster: suite.cluster,
jobScheduler: suite.jobScheduler, jobScheduler: suite.jobScheduler,