[jira] [Commented] (HBASE-4391) Add ability to start RS as root and call mlockall

["Todd Lipcon (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HBASE-4391?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13104074#comment-13104074 ] 

Todd Lipcon commented on HBASE-4391:
------------------------------------

I've put an initial implementation of a JVMTI agent that accomplishes this here:

https://github.com/toddlipcon/mlockall_agent

It can be compiled using the "compile.sh" script in the repo, and then configured like so in hbase-env.sh:
{code}
export HBASE_REGIONSERVER_OPTS="$HBASE_REGIONSERVER_OPTS -agentpath:/home/todd/git/libmlockall_agent/libmlockall_agent.so=user=todd"
{code}

After which point you must run the regionserver as root:
{code}

todd@todd-w510:~/git/hbase$ ./bin/hbase regionserver 
mlockall_agent: Unable to boost memlock resource limit: Operation not permitted
mlockall_agent: (this may be because java was not run as root!)
Error occurred during initialization of VM
agent library failed to init: /home/todd/git/libmlockall_agent/libmlockall_agent.so
todd@todd-w510:~/git/hbase$ sudo ./bin/hbase regionserver 
mlockall_agent: Successfully locked memory and setuid to todd
{code}

Perhaps some other people can check this out, and if it looks generally useful we could either build it along with HBase or at least point to it from docs?

> Add ability to start RS as root and call mlockall
> -------------------------------------------------
>
>                 Key: HBASE-4391
>                 URL: https://issues.apache.org/jira/browse/HBASE-4391
>             Project: HBase
>          Issue Type: New Feature
>          Components: regionserver
>    Affects Versions: 0.94.0
>            Reporter: Todd Lipcon
>            Assignee: Todd Lipcon
>             Fix For: 0.94.0
>
>
> A common issue we've seen in practice is that users oversubscribe their region servers with too many MR tasks, etc. As soon as the machine starts swapping, the RS grinds to a halt, loses ZK session, aborts, etc.
> This can be combatted by starting the RS as root, calling mlockall(), and then setuid down to the hbase user. We should not require this, but we should provide it as an option.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira