[ 
https://issues.apache.org/jira/browse/HBASE-10293?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13866180#comment-13866180
 ] 

Nick Dimiduk commented on HBASE-10293:
--------------------------------------

That's probably true for someone with production-level automation 
infrastructure. I'm thinking of the fellow who downloads a tarball, follows the 
instructions in the hbase-env.sh comments and is surprised by the outcome. 
FWIW, I don't think it's common to run each process out of it's own config 
directory. Likewise, I don't think it's common to set each process to log to 
it's own log directory either. Rather I tend to see /var/log/hbase containing 
all the HBase process logs for the machine.

*shrug* I don't feel strongly about the issue, it just surprised me while I was 
setting up some performance infra recently. If you'd prefer to defer this kind 
of concern to the puppetiers, I guess resolve as not a problem.

> Master and RS GC logs can conflict when run on same host
> --------------------------------------------------------
>
>                 Key: HBASE-10293
>                 URL: https://issues.apache.org/jira/browse/HBASE-10293
>             Project: HBase
>          Issue Type: Bug
>          Components: scripts
>    Affects Versions: 0.96.1.1
>            Reporter: Nick Dimiduk
>
> My issue manifests when I uncomment the line {{export SERVER_GC_OPTS=...}} in 
> hbase-env.sh and start HBase. It's a single node in distributed mode, so both 
> a Master and RegionServer are started on that host. Both start commands are 
> run in the same minute, so only one gc.log-`date` file is created. `lsof` 
> indicates two processes are writing to that file and the output of `ps` 
> confirms they both received the same 
> {{-Xloggc:/grid/0/var/log/hbase/gc.log-201401071515}} argument.
> Presumably, the same will happen for folks running the thrift and rest 
> gateways on the same box (any java process itemized in the server_cmds array 
> in bin/hbase).
> Related (the reason I discovered this issue in the first place), stopping the 
> master process results in its gc.log being truncated.



--
This message was sent by Atlassian JIRA
(v6.1.5#6160)

Reply via email to