Fix: data partition decommission timeout

Fix ths issue that Master report timeout error when executing data partition decommision operation. Signed-off-by: Mofei Zhang <[email protected]>
cubefs · Oct 10, 2020 · fffad00 · fffad00
1 parent d259d3f
commit fffad00
Showing 1 changed file with 17 additions and 5 deletions.
diff --git a/datanode/partition_raftfsm.go b/datanode/partition_raftfsm.go
@@ -17,12 +17,13 @@ package datanode
 import (
 	"encoding/json"
 	"fmt"
-	"github.com/chubaofs/chubaofs/storage"
 	"net"
+	"sync"
 	"sync/atomic"
 	"time"
 
 	"github.com/chubaofs/chubaofs/proto"
+	"github.com/chubaofs/chubaofs/storage"
 	"github.com/chubaofs/chubaofs/util/log"
 	"github.com/tiglabs/raft"
 	raftproto "github.com/tiglabs/raft/proto"
@@ -55,10 +56,21 @@ func (dp *DataPartition) ApplyMemberChange(confChange *raftproto.ConfChange, ind
 		}
 		isUpdated, err = dp.addRaftNode(req, index)
 		if isUpdated && err == nil {
-			dp.updateReplicas(true)
-			if dp.isLeader {
-				dp.ExtentStore().MoveAllToBrokenTinyExtentC(storage.TinyExtentCount)
-			}
+			// Perform the update replicas operation asynchronously after the execution of the member change applying
+			// related process.
+			updateWG := sync.WaitGroup{}
+			updateWG.Add(1)
+			defer updateWG.Done()
+			go func() {
+				updateWG.Wait()
+				if err = dp.updateReplicas(true); err != nil {
+					log.LogErrorf("ApplyMemberChange: update partition %v replicas failed: %v", dp.partitionID, err)
+					return
+				}
+				if dp.isLeader {
+					dp.ExtentStore().MoveAllToBrokenTinyExtentC(storage.TinyExtentCount)
+				}
+			}()
 		}
 	case raftproto.ConfRemoveNode:
 		req := &proto.RemoveDataPartitionRaftMemberRequest{}