[
https://issues.apache.org/jira/browse/HBASE-9721?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13853643#comment-13853643
]
Jimmy Xiang commented on HBASE-9721:
------------------------------------
After RS accepts the RPC, should the RS check both the znode version and data
before open the region? If so, we can avoid the problem without change the RPC,
right?
> RegionServer should not accept regionOpen RPC intended for another(previous)
> server
> -----------------------------------------------------------------------------------
>
> Key: HBASE-9721
> URL: https://issues.apache.org/jira/browse/HBASE-9721
> Project: HBase
> Issue Type: Bug
> Affects Versions: 0.98.0, 0.96.0
> Reporter: Enis Soztutar
> Assignee: Enis Soztutar
> Attachments: hbase-9721_v0.patch, hbase-9721_v1.patch,
> hbase-9721_v2.patch
>
>
> On a test cluster, this following events happened with ITBLL and CM leading
> to meta being unavailable until master is restarted.
> An RS carrying meta died, and master assigned the region to one of the RSs.
> {code}
> 2013-10-03 23:30:06,611 INFO
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-1]
> master.AssignmentManager: Assigning hbase:meta,,1.1588230740 to
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820
> 2013-10-03 23:30:06,611 INFO
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-1]
> master.RegionStates: Transitioned {1588230740 state=OFFLINE,
> ts=1380843006601, server=null} to {1588230740 state=PENDING_OPEN,
> ts=1380843006611,
> server=gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820}
> 2013-10-03 23:30:06,611 DEBUG
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-1]
> master.ServerManager: New admin connection to
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820
> {code}
> At the same time, the RS that meta recently got assigned also died (due to
> CM), and restarted:
> {code}
> 2013-10-03 23:30:07,636 DEBUG [RpcServer.handler=17,port=60000]
> master.ServerManager: REPORT: Server
> gs-hdp2-secure-1380781860-hbase-8.cs1cloud.internal,60020,1380843002494 came
> back up, removed it from the dead servers list
> 2013-10-03 23:30:08,769 INFO [RpcServer.handler=18,port=60000]
> master.ServerManager: Triggering server recovery; existingServer
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820 looks
> stale, new
> server:gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380843006362
> 2013-10-03 23:30:08,771 DEBUG [RpcServer.handler=18,port=60000]
> master.AssignmentManager: Checking region=hbase:meta,,1.1588230740, zk
> server=gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820
>
> current=gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820,
> matches=true
> 2013-10-03 23:30:08,771 DEBUG [RpcServer.handler=18,port=60000]
> master.ServerManager:
> Added=gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820
> to dead servers, submitted shutdown handler to be executed meta=true
> 2013-10-03 23:30:08,771 INFO [RpcServer.handler=18,port=60000]
> master.ServerManager: Registering
> server=gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380843006362
> 2013-10-03 23:30:08,772 INFO
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-2]
> handler.MetaServerShutdownHandler: Splitting hbase:meta logs for
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820
> {code}
> AM/SSH sees that the RS that died was carrying meta, but the assignment RPC
> request was still not sent:
> {code}
> 2013-10-03 23:30:08,791 DEBUG
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-2]
> master.AssignmentManager: Checking region=hbase:meta,,1.1588230740, zk
> server=gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820
>
> current=gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820,
> matches=true
> 2013-10-03 23:30:08,791 INFO
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-2]
> handler.MetaServerShutdownHandler: Server
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820 was
> carrying META. Trying to assign.
> 2013-10-03 23:30:08,791 DEBUG
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-2]
> master.RegionStates: Offline 1588230740 with current state=PENDING_OPEN,
> expected state=OFFLINE/SPLITTING/MERGING
> 2013-10-03 23:30:08,791 INFO
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-2]
> master.RegionStates: Transitioned {1588230740 state=PENDING_OPEN,
> ts=1380843006611,
> server=gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820}
> to {1588230740 state=OFFLINE, ts=1380843008791, server=null}
> 2013-10-03 23:30:09,809 INFO
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-2]
> zookeeper.ZooKeeperNodeTracker: Unsetting hbase:meta region location in
> ZooKeeper
> {code}
> Our first attempt at the assign rpc fails, because the new server is now
> starting. The second attempt though succeeds:
> {code}
> 2013-10-03 23:30:10,621 INFO
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-1]
> master.AssignmentManager: Assigning hbase:meta,,1.1588230740 to
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820
> 2013-10-03 23:30:10,621 INFO
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-1]
> master.RegionStates: Transitioned {1588230740 state=OFFLINE,
> ts=1380843008791, server=null} to {1588230740 state=PENDING_OPEN,
> ts=1380843010621,
> server=gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820}
> 2013-10-03 23:30:10,621 DEBUG
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-1]
> master.ServerManager: New admin connection to
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820
> 2013-10-03 23:30:10,622 DEBUG [RpcServer.handler=22,port=60000]
> master.ServerManager: REPORT: Server
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380843006362 came
> back up, removed it from the dead servers list
> 2013-10-03 23:30:10,934 DEBUG
> [MASTER_META_SERVER_OPERATIONS-gs-hdp2-secure-1380781860-hbase-12:60000-2]
> master.AssignmentManager: Skip assigning {ENCODED => 1588230740, NAME =>
> 'hbase:meta,,1', STARTKEY => '', ENDKEY => ''}, it is already {1588230740
> state=PENDING_OPEN, ts=1380843010621,
> server=gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820}
> {code}
> Note that the start time for the server does not match (1380842900820 is old
> , 1380843006362 is new).
> The region server got the rpc to open the region, but failed to change the zk
> state, because ServerNames is not matching:
> {code}
> 2013-10-03 23:30:10,601 INFO [Priority.RpcServer.handler=0,port=60020]
> regionserver.HRegionServer: Open hbase:meta,,1.1588230740
> 2013-10-03 23:30:10,897 DEBUG
> [RS_OPEN_META-gs-hdp2-secure-1380781860-hbase-5:60020-0] zookeeper.ZKAssign:
> regionserver:60020-0x1417d489d9b0bd6 Transitioning 1588230740 from
> M_ZK_REGION_OFFLINE to RS_ZK_REGION_OPENING
> 2013-10-03 23:30:10,918 WARN
> [RS_OPEN_META-gs-hdp2-secure-1380781860-hbase-5:60020-0] zookeeper.ZKAssign:
> regionserver:60020-0x1417d489d9b0bd6 Attempt to transition the unassigned
> node for 1588230740 from M_ZK_REGION_OFFLINE to RS_ZK_REGION_OPENING failed,
> the server that tried to transition was
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380843006362 not
> the expected
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820
> 2013-10-03 23:30:10,918 WARN
> [RS_OPEN_META-gs-hdp2-secure-1380781860-hbase-5:60020-0]
> handler.OpenRegionHandler: Failed transition from OFFLINE to OPENING for
> region=1588230740
> 2013-10-03 23:30:10,919 WARN
> [RS_OPEN_META-gs-hdp2-secure-1380781860-hbase-5:60020-0]
> handler.OpenRegionHandler: Region was hijacked? Opening cancelled for
> encodedName=1588230740
> 2013-10-03 23:30:10,919 INFO
> [RS_OPEN_META-gs-hdp2-secure-1380781860-hbase-5:60020-0]
> handler.OpenRegionHandler: Opening of region {ENCODED => 1588230740, NAME =>
> 'hbase:meta,,1', STARTKEY => '', ENDKEY => ''} failed, transitioning from
> OFFLINE to FAILED_OPEN in ZK, expecting version 0
> 2013-10-03 23:30:10,919 DEBUG
> [RS_OPEN_META-gs-hdp2-secure-1380781860-hbase-5:60020-0] zookeeper.ZKAssign:
> regionserver:60020-0x1417d489d9b0bd6 Transitioning 1588230740 from
> M_ZK_REGION_OFFLINE to RS_ZK_REGION_FAILED_OPEN
> 2013-10-03 23:30:10,921 WARN
> [RS_OPEN_META-gs-hdp2-secure-1380781860-hbase-5:60020-0] zookeeper.ZKAssign:
> regionserver:60020-0x1417d489d9b0bd6 Attempt to transition the unassigned
> node for 1588230740 from M_ZK_REGION_OFFLINE to RS_ZK_REGION_FAILED_OPEN
> failed, the server that tried to transition was
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380843006362 not
> the expected
> gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820
> 2013-10-03 23:30:10,921 WARN
> [RS_OPEN_META-gs-hdp2-secure-1380781860-hbase-5:60020-0]
> handler.OpenRegionHandler: Unable to mark region {ENCODED => 1588230740, NAME
> => 'hbase:meta,,1', STARTKEY => '', ENDKEY => ''} as FAILED_OPEN. It's likely
> that the master already timed out this open attempt, and thus another RS
> already has the region.
> {code}
> It seems that the RS behaved correct by not being able to open the region by
> transitioning the zk assignment node. However, the master fails to timeout
> the assignment even though the meta region is reported in RIT:
> {code}
> 2013-10-04 00:14:50,658 DEBUG
> [gs-hdp2-secure-1380781860-hbase-12.cs1cloud.internal,60000,1380842679864-BalancerChore]
> master.HMaster: Not running balancer because 1 region(s) in transition:
> {1588230740={1588230740 state=PENDING_OPEN, ts=1380843010621,
> server=gs-hdp2-secure-1380781860-hbase-5.cs1cloud.internal,60020,1380842900820}}
> {code}
--
This message was sent by Atlassian JIRA
(v6.1.4#6159)
