hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Nick Dimiduk (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HBASE-14237) Meta region may be onlined on multi regonservers for bugs of assigning meta
Date Tue, 18 Aug 2015 19:12:45 GMT

    [ https://issues.apache.org/jira/browse/HBASE-14237?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14701808#comment-14701808

Nick Dimiduk commented on HBASE-14237:

Hi [~liushaohui]. I have to ask the question -- any chance you can upgrade to a 0.98+ release?
If I remember the history correctly, our Chaos Monkey test harness was introduced after 0.94.
Many bugs in AM have been found and fixed via this testing on newer branches. We've back ported
what's been possible, but at this point the AM in 0.94 is very different from more recent

> Meta region may be onlined on multi regonservers for bugs of assigning meta
> ---------------------------------------------------------------------------
>                 Key: HBASE-14237
>                 URL: https://issues.apache.org/jira/browse/HBASE-14237
>             Project: HBase
>          Issue Type: Bug
>    Affects Versions: 0.94.11
>            Reporter: Liu Shaohui
>            Assignee: Liu Shaohui
>            Priority: Critical
>         Attachments: meta.log
> When a regionserver failed to open the meta region and crash after setting the RS_ZK_REGION_FAILED_OPEN
state of meta region in zookeeper, the master will handle the event of RS_ZK_REGION_FAILED_OPEN
and try to assign the meta region again in AssignmentManager#handleRegion. But at the same
time, the master will handle the regionserver expired event and start a MetaServerShutdownHandler
for the regionserver, because the servername of regionserver is same as the servername of
the unassigned node of meta region. In the MetaServerShutdownHandler, the meta region may
be assigned for second time.
> [~heliangliang]
> We have encountered this problem in our production cluster which resulted in inconsistency
of region location in meta table. You can see the log from the attachment.
> The code of AssignmentManager is so complex and I have not get a solution to fix this
problem. Could someone kindly help to give some suggestions? Thanks

This message was sent by Atlassian JIRA

View raw message