[HDFS-12132] Both two NameNodes become Standby because the ZKFC exception

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.8.1
• Component/s: auto-failover
• Assignee: Unassigned
• Reporter: Jiandan Yang
• Created: 13/Jul/17 08:13
• Updated: 03/Feb/18 03:47

Description

Active NameNode become Standby because the ZKFC exception and Standby NameNode is still Standby When rolling upgrading Hadoop from Hadoop-2.6.5 to Hadoop-2.8.0, this lead HDFS to be not available. The logic of processing exception in ZKFC seems to be problematic, ZKFC should guarantee to have a active NameNode.

Before upgrading, the cluster was deployed with HA, NN1 was active, and NN2 was standby
The configuration before upgrading is as follows:

dfs.namenode.rpc-address.nameservice.nn1 nn1: 8020
dfs.namenode.rpc-address.nameservice.nn2 nn2: 8020

After upgrading, add the configuration of the separate RPC service:

dfs.namenode.rpc-address.nameservice.nn1 nn1: 8020
dfs.namenode.rpc-address.nameservice.nn2 nn2: 8020
dfs.namenode.servicerpc-address.nameservice.nn1 nn1: 8021
dfs.namenode.servicerpc-address.nameservice.nn2 nn2: 8021
dfs.namenode.lifeline.rpc-address.nameservice.nn1 nn1: 8022
dfs.namenode.lifeline.rpc-address.nameservice.nn2 nn2: 8022

The upgrade steps are as follows:
1. Upgrade NN2: restart NameNode process on NN2
2. Upgrade NN1: restart the NameNode process on NN1, then NN2 becomes active, NN1 is standby
3. Restart both ZKFC on NN1 and NN2

After restarting ZKFC, Active NameNodes have become Standby, and Standby NameNode did not become Active. Two ZKFC having been doing a loop and threw many same exceptions.

createLockNodeAsync()  // create lock successfully
becomeActive()  // return false
terminateConnection()  // delete EPHEMERAL znode of 'ActiveStandbyElectorLock'  
sleepFor(sleepTime)

After running command 'hdfs zkfc -formatZK', ZKFC backs to normal.
ZKFC Exception log is:

2017-07-11 18:49:44,311 WARN [main-EventThread] org.apache.hadoop.ha.ActiveStandbyElector: Exception handling the winning of election
java.lang.RuntimeException: Mismatched address stored in ZK for NameNode at nn2/xx.xxx.xx.xxx:8022: Stored protobuf was nameserviceId: “nameservice”
namenodeId: "nn2"
hostname: “nn2_hostname”
port: 8020
zkfcPort: 8019
, address from our own configuration for this NameNode was nn2_hostname/xx.xxx.xx.xxx:8021
        at org.apache.hadoop.hdfs.tools.DFSZKFailoverController.dataToTarget(DFSZKFailoverController.java:87)
        at org.apache.hadoop.ha.ZKFailoverController.fenceOldActive(ZKFailoverController.java:506)
        at org.apache.hadoop.ha.ZKFailoverController.access$1100(ZKFailoverController.java:61)
        at org.apache.hadoop.ha.ZKFailoverController$ElectorCallbacks.fenceOldActive(ZKFailoverController.java:895)
        at org.apache.hadoop.ha.ActiveStandbyElector.fenceOldActive(ActiveStandbyElector.java:985)
        at org.apache.hadoop.ha.ActiveStandbyElector.becomeActive(ActiveStandbyElector.java:882)
        at org.apache.hadoop.ha.ActiveStandbyElector.processResult(ActiveStandbyElector.java:467)
        at org.apache.zookeeper.ClientCnxn$EventThread.processEvent(ClientCnxn.java:599)
        at org.apache.zookeeper.ClientCnxn$EventThread.run(ClientCnxn.java:498)
2017-07-11 18:49:44,311 INFO [main-EventThread] org.apache.hadoop.ha.ActiveStandbyElector: Trying to re-establish ZK session
2017-07-11 18:49:44,311 INFO [main-EventThread] org.apache.zookeeper.ZooKeeper: Session: 0x15c3ada0ec319aa closed