Fix: refresh leader status, prevent master from getting an `Unavailab…

…le` leader for data partition

Signed-off-by: xuxihao <[email protected]>

master/data_partition_check.go

@@ -73,11 +73,24 @@ func (partition *DataPartition) checkReplicaStatusOnLiveNode(liveReplicas []*Dat
 }
 
 func (partition *DataPartition) checkReplicaStatus(timeOutSec int64) {
-	partition.RLock()
-	defer partition.RUnlock()
+	partition.Lock()
+	defer partition.Unlock()
 	for _, replica := range partition.Replicas {
-		replica.isLive(timeOutSec)
+		if !replica.isLive(timeOutSec) {
+			replica.Status = proto.Unavailable
+		}
+	}
+}
+
+func (partition *DataPartition) checkLeader(timeOut int64) {
+	partition.Lock()
+	defer partition.Unlock()
+	for _, dr := range partition.Replicas {
+		if !dr.isLive(timeOut) {
+			dr.IsLeader = false
+		}
 	}
+	return
 }
 
 // Check if there is any missing replica for a data partition.

master/vol.go

@@ -227,7 +227,7 @@ func (vol *Vol) checkDataPartitions(c *Cluster) (cnt int) {
 	for _, dp := range vol.dataPartitions.partitionMap {
 		dp.checkReplicaStatus(c.cfg.DataPartitionTimeOutSec)
 		dp.checkStatus(c.Name, true, c.cfg.DataPartitionTimeOutSec)
-
+		dp.checkLeader(c.cfg.DataPartitionTimeOutSec)
 		dp.checkMissingReplicas(c.Name, c.leaderInfo.addr, c.cfg.MissingDataPartitionInterval, c.cfg.IntervalToAlarmMissingDataPartition)
 		dp.checkReplicaNum(c, vol)
 		if dp.Status == proto.ReadWrite {