hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "stack (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HBASE-2482) regions in transition do not get reassigned by master when RS crashes
Date Mon, 03 May 2010 02:19:56 GMT

     [ https://issues.apache.org/jira/browse/HBASE-2482?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel

stack updated HBASE-2482:

    Attachment: 2482-unittest.txt

First cut at unit test.  Needs an edit but looks to be working.  Adds a protected 'killl'
to regionserver which simulates RS kill (does no cleanup but shutdown of socket).  Also added
new HMsg called TEST_MSG_BLOCK_RS.  When RS receives this from master, it just waits until
closed, aborted or killed.  It blocks the worker queue.

The way the test works is that it adds a new RS to small cluster, waits on load balancer to
move some regions to new server.  As soon as some are open, we send a close of them all followed
by a TEST_MSG_BLOCK_RS.  The closes go through, balancer assigns the new server some of the
closed regions only the TEST_MSG_BLOCK_RS is in place.

Let me make a patch that includes Todds patch and cleaned up test next.

> regions in transition do not get reassigned by master when RS crashes
> ---------------------------------------------------------------------
>                 Key: HBASE-2482
>                 URL: https://issues.apache.org/jira/browse/HBASE-2482
>             Project: Hadoop HBase
>          Issue Type: Bug
>          Components: master
>    Affects Versions: 0.20.5
>            Reporter: Todd Lipcon
>            Assignee: Todd Lipcon
>            Priority: Blocker
>             Fix For: 0.20.5, 0.21.0
>         Attachments: 2482-unittest.txt, hbase-2482.txt
> Very similar to HBASE-1928, but for the general case (not just ROOT/META):
> If a region is in transition on a RS when the RS crashes, the master does not remove
it from regionsInTransition when processing the RS shutdown. This is fairly easy to trigger
by bringing up a RS and kill -9ing it just as it starts to get regions assigned. Those regions
will get permanently lost since they're stuck in regionsInTransition and thus don't get assigned
by the metascanner.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message