fix: delegator filter out all partition's delete msg when loading segment (#31585)

May cause deleted data queryable a period of time.
relate: https://github.com/milvus-io/milvus/issues/31484
https://github.com/milvus-io/milvus/issues/31548

---------

Signed-off-by: aoiasd <zhicheng.yue@zilliz.com>

internal/querynodev2/delegator/delegator_data.go

@@ -749,7 +749,7 @@ func (sd *shardDelegator) readDeleteFromMsgstream(ctx context.Context, position
 			for _, tsMsg := range msgPack.Msgs {
 				if tsMsg.Type() == commonpb.MsgType_Delete {
 					dmsg := tsMsg.(*msgstream.DeleteMsg)
-					if dmsg.CollectionID != sd.collectionID || dmsg.GetPartitionID() != candidate.Partition() {
+					if dmsg.CollectionID != sd.collectionID || (dmsg.GetPartitionID() != common.AllPartitionsID && dmsg.GetPartitionID() != candidate.Partition()) {
 						continue
 					}
 

internal/querynodev2/delegator/delegator_data_test.go

@@ -1114,6 +1114,39 @@ func (s *DelegatorDataSuite) TestLevel0Deletions() {
 	s.True(pks[0].EQ(allPartitionDeleteData.Pks[0]))
 }
 
+func (s *DelegatorDataSuite) TestReadDeleteFromMsgstream() {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	s.mq.EXPECT().AsConsumer(mock.Anything, mock.Anything, mock.Anything, mock.Anything).Return(nil)
+	s.mq.EXPECT().Seek(mock.Anything, mock.Anything).Return(nil)
+	s.mq.EXPECT().Close()
+	ch := make(chan *msgstream.MsgPack, 10)
+	s.mq.EXPECT().Chan().Return(ch)
+
+	oracle := pkoracle.NewBloomFilterSet(1, 1, commonpb.SegmentState_Sealed)
+	oracle.UpdateBloomFilter([]storage.PrimaryKey{storage.NewInt64PrimaryKey(1), storage.NewInt64PrimaryKey(2)})
+
+	baseMsg := &commonpb.MsgBase{MsgType: commonpb.MsgType_Delete}
+
+	datas := []*msgstream.MsgPack{
+		{EndTs: 10, EndPositions: []*msgpb.MsgPosition{{Timestamp: 10}}, Msgs: []msgstream.TsMsg{
+			&msgstream.DeleteMsg{DeleteRequest: msgpb.DeleteRequest{Base: baseMsg, CollectionID: s.collectionID, PartitionID: 1, PrimaryKeys: storage.ParseInt64s2IDs(1), Timestamps: []uint64{1}}},
+			&msgstream.DeleteMsg{DeleteRequest: msgpb.DeleteRequest{Base: baseMsg, CollectionID: s.collectionID, PartitionID: -1, PrimaryKeys: storage.ParseInt64s2IDs(2), Timestamps: []uint64{5}}},
+			// invalid msg because partition wrong
+			&msgstream.DeleteMsg{DeleteRequest: msgpb.DeleteRequest{Base: baseMsg, CollectionID: s.collectionID, PartitionID: 2, PrimaryKeys: storage.ParseInt64s2IDs(1), Timestamps: []uint64{10}}},
+		}},
+	}
+
+	for _, data := range datas {
+		ch <- data
+	}
+
+	result, err := s.delegator.readDeleteFromMsgstream(ctx, &msgpb.MsgPosition{Timestamp: 0}, 10, oracle)
+	s.NoError(err)
+	s.Equal(2, len(result.Pks))
+}
+
 func TestDelegatorDataSuite(t *testing.T) {
 	suite.Run(t, new(DelegatorDataSuite))
 }

internal/storage/primary_key.go

@@ -410,3 +410,18 @@ func ParsePrimaryKeys2IDs(pks []PrimaryKey) *schemapb.IDs {
 
 	return ret
 }
+
+func ParseInt64s2IDs(pks ...int64) *schemapb.IDs {
+	ret := &schemapb.IDs{}
+	if len(pks) == 0 {
+		return ret
+	}
+
+	ret.IdField = &schemapb.IDs_IntId{
+		IntId: &schemapb.LongArray{
+			Data: pks,
+		},
+	}
+
+	return ret
+}