[
https://issues.apache.org/jira/browse/HADOOP-2276?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jim Kellerman updated HADOOP-2276:
----------------------------------
Attachment: patch.txt
Added synchronization to prevent multiple assignment of a region.
Modified ProcessServerShutdown to requeue itself on the delay queue if all the
meta regions are not online. This prevents spinning resulting from
requeue/dequeue on the normal message queue and allows other threads to get the
meta regions back on line
> [hbase] HADOOP-2274 introduces regression
> -----------------------------------------
>
> Key: HADOOP-2276
> URL: https://issues.apache.org/jira/browse/HADOOP-2276
> Project: Hadoop
> Issue Type: Bug
> Components: contrib/hbase
> Affects Versions: 0.16.0
> Reporter: Jim Kellerman
> Assignee: Jim Kellerman
> Fix For: 0.16.0
>
> Attachments: patch.txt
>
>
> HADOOP-2274 introduced race conditions that cause a regression which was the
> reason for Hadoop-Nightly build #314's failure.
> Symptoms of this regression are:
> - Assigning a single region to multiple region servers.
> - Thread starvation resulting in server lease expiration when, in fact, the
> region server had reported in but the main processing thread in the master
> did not renew the server's lease soon enough.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
