hadoop-common-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jim Kellerman (JIRA)" <j...@apache.org>
Subject [jira] Updated: (HADOOP-1960) [hbase] If a region server cannot talk to the master after several attempts, it should shut itself down
Date Tue, 02 Oct 2007 19:23:50 GMT

     [ https://issues.apache.org/jira/browse/HADOOP-1960?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Jim Kellerman updated HADOOP-1960:
----------------------------------

    Attachment: patch.txt

This patch removes the new unit test for testing region server shutdown because the change
to HMaster (adding an abort() method), is too dangerous to leave enabled. Without the test,
changes to HMaster and MiniHBase cluster are no longer needed.

Changes included in this patch are:

TestRegionServerAbort

- Add check for scanner != null before trying to close it

TestSplit

- Enclose test body in try catch block so that exceptions can be dumped to the console at
the point in the test where they occur.

HRegionServer

- If unable to communicate with the master for more than the lease timeout interval abort
server.



> [hbase] If a region server cannot talk to the master after several attempts, it should
shut itself down
> -------------------------------------------------------------------------------------------------------
>
>                 Key: HADOOP-1960
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1960
>             Project: Hadoop
>          Issue Type: Improvement
>          Components: contrib/hbase
>    Affects Versions: 0.15.0
>            Reporter: Jim Kellerman
>            Assignee: Jim Kellerman
>             Fix For: 0.15.0
>
>         Attachments: patch.txt, patch.txt
>
>
> If a region server cannot contact the master after a configurable number of tries, it
should shut itself down.
> If the region server cannot contact the master,
> - if the master is alive but the network is partitioned, the master will probably time
out the region server's lease and try to recover the server's log and reassign the regions
the server is serving.
> - if the master has died, and subsequently restarts, it will be reassigning regions anyway,
so the region server should stop serving the regions.

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


Mime
View raw message