[ 
https://issues.apache.org/jira/browse/HADOOP-1527?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12521516
 ] 

Jim Kellerman commented on HADOOP-1527:
---------------------------------------

This really is an abnormal condition, because if a region server dies, the 
master should split the region server's log (and place the records in the 
regions' directory(ies)) and then remove the region server log.

If a region server is starting up and discovers a log directory exists which 
should belong exclusively to that server that means that either:
- the master has not cleaned up the log yet (or perhaps never will if the 
master crashed before it could)
- another region server started and grabbed that port, so the starting region 
server should shut down.

In the former case, if the master crashed, we should provide a tool that can 
split the log so we can recover the regions that the previous region server 
instance was serving.

Otherwise I think that what is happening is the correct behavior.


> Region server won't start because logdir exists
> -----------------------------------------------
>
>                 Key: HADOOP-1527
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1527
>             Project: Hadoop
>          Issue Type: Bug
>          Components: contrib/hbase
>            Reporter: stack
>            Assignee: Jim Kellerman
>
> Starting and then impolitely stopping a cluster I came across the following:
> 2007-06-25 19:43:31,449 ERROR org.apache.hadoop.hbase.HRegionServer: Can not 
> start region server because 
> org.apache.hadoop.hbase.RegionServerRunningException: region server already 
> running at 208.76.44.140:60010 because logdir  exists
>         at 
> org.apache.hadoop.hbase.HRegionServer.<init>(HRegionServer.java:447)
>         at 
> org.apache.hadoop.hbase.HRegionServer.<init>(HRegionServer.java:372)
>         at org.apache.hadoop.hbase.HRegionServer.main(HRegionServer.java:1233)
> Region server should recover or offer a recovery path when we run into this 
> condition.

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

Reply via email to