fix: Delegator stuck at unserviceable status (#37694)

issue: #37679

pr #36549 introduce the logic error which update current target when
only parts of channel is ready.

This PR fix the logic error and let dist handler keep pull distribution
on querynode until all delegator becomes serviceable.

Signed-off-by: Wei Liu <wei.liu@zilliz.com>

internal/querycoordv2/dist/dist_handler.go

@@ -38,6 +38,7 @@ import (
 	"github.com/milvus-io/milvus/pkg/util/commonpbutil"
 	"github.com/milvus-io/milvus/pkg/util/merr"
 	"github.com/milvus-io/milvus/pkg/util/paramtable"
+	"github.com/milvus-io/milvus/pkg/util/typeutil"
 )
 
 type TriggerUpdateTargetVersion = func(collectionID int64)
@@ -195,6 +196,8 @@ func (dh *distHandler) updateLeaderView(resp *querypb.GetDataDistributionRespons
 	channels := lo.SliceToMap(resp.GetChannels(), func(channel *querypb.ChannelVersionInfo) (string, *querypb.ChannelVersionInfo) {
 		return channel.GetChannel(), channel
 	})
+
+	collectionsToSync := typeutil.NewUniqueSet()
 	for _, lview := range resp.GetLeaderViews() {
 		segments := make(map[int64]*meta.Segment)
 
@@ -247,9 +250,10 @@ func (dh *distHandler) updateLeaderView(resp *querypb.GetDataDistributionRespons
 			err := merr.WrapErrServiceInternal(fmt.Sprintf("target version mismatch, collection: %d, channel: %s,  current target version: %v, leader version: %v",
 				lview.GetCollection(), lview.GetChannel(), currentTargetVersion, lview.TargetVersion))
 
-			// segment and channel already loaded, trigger target observer to check target version
-			dh.syncTargetVersionFn(lview.GetCollection())
 			view.UnServiceableError = err
+			// make dist handler pull next distribution until all delegator is serviceable
+			dh.lastUpdateTs = 0
+			collectionsToSync.Insert(lview.Collection)
 			log.Info("leader is not available due to target version not ready",
 				zap.Int64("collectionID", view.CollectionID),
 				zap.Int64("nodeID", view.ID),
@@ -259,6 +263,12 @@ func (dh *distHandler) updateLeaderView(resp *querypb.GetDataDistributionRespons
 	}
 
 	dh.dist.LeaderViewManager.Update(resp.GetNodeID(), updates...)
+
+	// segment and channel already loaded, trigger target observer to update
+	collectionsToSync.Range(func(collection int64) bool {
+		dh.syncTargetVersionFn(collection)
+		return true
+	})
 }
 
 func (dh *distHandler) getDistribution(ctx context.Context) (*querypb.GetDataDistributionResponse, error) {

internal/querycoordv2/observers/target_observer.go

@@ -386,16 +386,16 @@ func (ob *TargetObserver) shouldUpdateCurrentTarget(ctx context.Context, collect
 		channelReadyLeaders := lo.Filter(ob.distMgr.LeaderViewManager.GetByFilter(meta.WithChannelName2LeaderView(channel)), func(leader *meta.LeaderView, _ int) bool {
 			return utils.CheckDelegatorDataReady(ob.nodeMgr, ob.targetMgr, leader, meta.NextTarget) == nil
 		})
-		collectionReadyLeaders = append(collectionReadyLeaders, channelReadyLeaders...)
 
 		// to avoid stuck here in dynamic increase replica case, we just check available delegator number
-		if int32(len(collectionReadyLeaders)) < replicaNum {
+		if int32(len(channelReadyLeaders)) < replicaNum {
 			log.RatedInfo(10, "channel not ready",
 				zap.Int("readyReplicaNum", len(channelReadyLeaders)),
 				zap.String("channelName", channel),
 			)
 			return false
 		}
+		collectionReadyLeaders = append(collectionReadyLeaders, channelReadyLeaders...)
 	}
 
 	var collectionInfo *milvuspb.DescribeCollectionResponse

tests/integration/target/target_test.go

@@ -147,7 +147,6 @@ func (s *TargetTestSuit) TestQueryCoordRestart() {
 	s.initCollection(name, 1, 2, 2, 2000)
 
 	ctx := context.Background()
-
 	info, err := s.Cluster.Proxy.DescribeCollection(ctx, &milvuspb.DescribeCollectionRequest{
 		Base:           commonpbutil.NewMsgBase(),
 		CollectionName: name,
@@ -156,6 +155,16 @@ func (s *TargetTestSuit) TestQueryCoordRestart() {
 	s.True(merr.Ok(info.GetStatus()))
 	collectionID := info.GetCollectionID()
 
+	// wait until all shards are ready
+	// cause showCollections won't just wait all collection becomes loaded, proxy will use retry to block until all shard are ready
+	s.Eventually(func() bool {
+		resp, err := s.Cluster.QueryCoord.GetShardLeaders(ctx, &querypb.GetShardLeadersRequest{
+			Base:         commonpbutil.NewMsgBase(),
+			CollectionID: collectionID,
+		})
+		return err == nil && merr.Ok(resp.GetStatus()) && len(resp.Shards) == 2
+	}, 60*time.Second, 1*time.Second)
+
 	// trigger old coord stop
 	s.Cluster.StopQueryCoord()