Actually I see now that this EOFException is keeping a region offline, are 
there anyways around this error to bring the region back online? I don't have 
the logs from the regionservers when it went offline but here is the section of 
the master log from then:

http://pastebin.com/4ZBKGbnZ

thanks again
-chris

On Mar 2, 2011, at 1:03 AM, Chris Tarnas wrote:

> Under heavy loads I've seen a few of EOFException errors in my regionserver 
> logs:
> 
> 2011-03-02 02:27:03,669 ERROR 
> org.apache.hadoop.hbase.regionserver.HRegionServer: Error opening 
> sequence,h7BpVjo07UDYrkBZBLwWfg\x09fc00fc97be11e00d731605f8e061462c-A2610001-1\x09,1298335975607.8a5d1e4a300792d74f516ba26de869c8.
> java.io.EOFException: 
> hdfs://lxbt006-pvt:8020/hbase/sequence/8a5d1e4a300792d74f516ba26de869c8/recovered.edits/0000000000054475364,
>  entryStart=2336278916, pos=2336278916, end=4672557832, edit=13370
>       at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)
>       at 
> sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:39)
>       at 
> sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:27)
> 
> Checking the same timeframe in the namenode logs on lcbt006-pvt reveals no 
> ominous messages (no warns, errors, anything), just the same file being 
> opened by a different node:
> 
> 2011-03-02 02:27:05,466 INFO 
> org.apache.hadoop.hdfs.server.namenode.FSNamesystem.audit: ugi=hadoop      
> ip=/10.56.24.13 cmd=open        
> src=/hbase/sequence/8a5d1e4a300792d74f516ba26de869c8/recovered.edits/0000000000054475364
>         dst=null        perm=null
> 
> 
> The Troubleshooting Wiki mentions it is related to swapping, but none of the 
> nodes are swapping - they all have plenty of RAM. Are there other common 
> causes? Is this anything I should be worried about or just "normal" 
> exceptions, anything else I should look for? I'm on cdh3b3 and will be moving 
> to b4 once I get a chance to run it through a test cluster.
> 
> thank you,
> -chris

Reply via email to