[HBASE-14664] Master failover issue: Backup master is unable to start if active master is killed and started in short time interval

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.0.0
• Fix Version/s: 2.0.0
• Component/s: master
• Assignee: Samir Ahmic
• Reporter: Samir Ahmic
• Created: 21/Oct/15 19:08
• Updated: 30/Nov/15 05:43
• Resolved: 27/Nov/15 13:48

Description

I notice this issue while running IntegrationTestDDLMasterFailover, it can be simply reproduced by executing this on active master (tested on two masters + 3rs cluster setup)

$ kill -9 master_pid; hbase-daemon.sh  start master

Logs show that new active master is trying to locate hbase:meta table on restarted active master

2015-10-21 19:28:20,804 INFO  [hnode2:16000.activeMasterManager] zookeeper.MetaTableLocator: Failed verification of hbase:meta,,1 at address=hnode1,16000,1445447051681, exception=org.apache.hadoop.hbase.ipc.ServerNotRunningYetException: Server is not running yet
        at org.apache.hadoop.hbase.regionserver.RSRpcServices.checkOpen(RSRpcServices.java:1092)
        at org.apache.hadoop.hbase.regionserver.RSRpcServices.getRegionInfo(RSRpcServices.java:1330)
        at org.apache.hadoop.hbase.master.MasterRpcServices.getRegionInfo(MasterRpcServices.java:1525)
        at org.apache.hadoop.hbase.protobuf.generated.AdminProtos$AdminService$2.callBlockingMethod(AdminProtos.java:22233)
        at org.apache.hadoop.hbase.ipc.RpcServer.call(RpcServer.java:2136)
        at org.apache.hadoop.hbase.ipc.CallRunner.run(CallRunner.java:106)
        at org.apache.hadoop.hbase.ipc.RpcExecutor.consumerLoop(RpcExecutor.java:130)
        at org.apache.hadoop.hbase.ipc.RpcExecutor$1.run(RpcExecutor.java:107)
        at java.lang.Thread.run(Thread.java:745)
2015-10-21 19:28:20,805 INFO  [hnode2:16000.activeMasterManager] master.HMaster: Meta was in transition on hnode1,16000,1445447051681
2015-10-21 19:28:20,805 INFO  [hnode2:16000.activeMasterManager] master.AssignmentManager: Processing {1588230740 state=OPEN, ts=1445448500598, server=hnode1,16000,1445447051681

and because of above master is unable to read hbase:meta table:

2015-10-21 19:28:49,429 INFO  [hconnection-0x6e9cebcc-shared--pool6-t1] client.AsyncProcess: #2, table=hbase:meta, attempt=10/351 failed=1ops, last exception: org.apache.hadoop.hbase.ipc.ServerNotRunningYetException: org.apache.hadoop.hbase.ipc.ServerNotRunningYetException: Server is not running yet
        at org.apache.hadoop.hbase.regionserver.RSRpcServices.checkOpen(RSRpcServices.java:1092)
        at org.apache.hadoop.hbase.regionserver.RSRpcServices.multi(RSRpcServices.java:2083)
        at org.apache.hadoop.hbase.protobuf.generated.ClientProtos$ClientService$2.callBlockingMethod(ClientProtos.java:32462)
        at org.apache.hadoop.hbase.ipc.RpcServer.call(RpcServer.java:2136)
        at org.apache.hadoop.hbase.ipc.CallRunner.run(CallRunner.java:106)
        at org.apache.hadoop.hbase.ipc.RpcExecutor.consumerLoop(RpcExecutor.java:130)
        at org.apache.hadoop.hbase.ipc.RpcExecutor$1.run(RpcExecutor.java:107)
        at java.lang.Thread.run(Thread.java:745)

which cause master is unable to complete start.
I have also notices that in this case value of /hbase/meta-region-server znode is always pointing on restarted active master (hnode1 in my cluster ).
I was able to workaround this issue by repeating same scenario with following:

$ kill -9 master_pid; hbase zkcli rmr /hbase/meta-region-server; hbase-daemon.sh start master

So issue is probably caused by staled value in /hbase/meta-region-server znode. I will try to create patch based on above.

Issue Links

• relates to: HBASE-14861 HBASE_ZNODE_FILE on master server is overwritten by regionserver process in case of master-rs collocation