[HBASE-12440] Region may remain offline on clean startup under certain race condition

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.98.8, 0.99.2
• Component/s: Region Assignment
• Hadoop Flags: Reviewed
• Assignee: Virag Kothari
• Reporter: Virag Kothari
• Created: 06/Nov/14 23:22
• Updated: 21/Feb/15 23:46
• Resolved: 07/Nov/14 23:40

Description

Saw this in prod some time back with zk assignment
On clean startup, while master was doing bulk assign while one of the region servers dies. The bulk assigner then tried to assign it individually using AssignCallable. The AssignCallable does a forceStateToOffline() and skips assigning as it wants the SSH to do the assignment

2014-10-16 16:05:23,593 DEBUG master.AssignmentManager [AM.-pool1-t1] : Offline sieve_main:inlinks,com.cbslocal.seattle/photo-galleries/category/consumer///:http\x09com.cbslocal.seattle/photo-galleries/category/tailgate-fan///:http,1413464068567.1f1620174d2542fe7d5b034f3311c3a8., no need to unassign since it's on a dead server: gsbl872n06.blue.ygrid.yahoo.com,50511,1413475494016
2014-10-16 16:05:23,593  INFO master.RegionStates [AM.-pool1-t1] : Transition {1f1620174d2542fe7d5b034f3311c3a8 state=PENDING_OPEN, ts=1413475519482, server=gsbl872n06.blue.ygrid.yahoo.com,50511,1413475494016} to {1f1620174d2542fe7d5b034f3311c3a8 state=OFFLINE, ts=1413475523593, server=gsbl872n06.blue.ygrid.yahoo.com,50511,1413475494016}
2014-10-16 16:05:23,598  INFO master.AssignmentManager [AM.-pool1-t1] : Skip assigning sieve_main:inlinks,com.cbslocal.seattle/photo-galleries/category/consumer///:http\x09com.cbslocal.seattle/photo-galleries/category/tailgate-fan///:http,1413464068567.1f1620174d2542fe7d5b034f3311c3a8., it is on a dead but not processed yet server: gsbl872n06.blue.ygrid.yahoo.com,50511,1413475494016

But the SSH wont assign as the region is offline but not in transition

2014-10-16 16:05:24,606  INFO handler.ServerShutdownHandler [MASTER_SERVER_OPERATIONS-hbbl874n38:50510-0] : Reassigning 0 region(s) that gsbl872n06.blue.ygrid.yahoo.com,50511,1413475494016 was carrying (and 0 regions(s) that were opening on this server)
2014-10-16 16:05:24,606 DEBUG master.DeadServer [MASTER_SERVER_OPERATIONS-hbbl874n38:50510-0] : Finished processing gsbl872n06.blue.ygrid.yahoo.com,50511,1413475494016

In zk-less assignment, the bulk assigner invoking AssignCallable and the SSH may try to assign the region. But as they go through lock, only one will succeed and doesn't seem to be an issue.