[
https://issues.apache.org/jira/browse/MAPREDUCE-2265?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12983262#action_12983262
]
Owen O'Malley commented on MAPREDUCE-2265:
------------------------------------------
{quote}
We already do this for the native libs, why is the task-controller any
different?
{quote}
Agreed, we need to push this change all of the way through, but you've chosen
to start here. I'm precisely arguing that all of the deployment directories
should assume a single architecture with the exception of 32/64 bit libraries
where you clearly need both. Does that sound reasonable? I think that users
would rather download a "x86 redhat" tarball than a tarball that includes all
possible combinations of hardware and operating system.
{quote}
the scripts never call the task-controller, only the TT does,
{quote}
The scripts calls jsvc and it should be in the same place as task-controller.
{quote}
Then the tarball environment and the product of a build start looking rather
different, we have extra install complexity, etc.
{quote}
Running out of the build directory isn't a good option until we undo the
project split anyways. Once we do, then we could make a single build directory
that contains a deployment directory.
> task-controller and jsvc should install into sbin/<platform>/ directory
> -----------------------------------------------------------------------
>
> Key: MAPREDUCE-2265
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-2265
> Project: Hadoop Map/Reduce
> Issue Type: Improvement
> Components: build, task-controller, tasktracker
> Affects Versions: 0.22.0
> Reporter: Todd Lipcon
> Assignee: Todd Lipcon
> Priority: Blocker
> Fix For: 0.22.0
>
>
> Currently the task-controller and jsvc "live" in the bin/ directory
> regardless of build platform. This is incorrect since these components are
> native compiled code and thus are built for a particular architecture. So,
> when we ship a build of 22, we will want to ship both 32-bit and 64-bit
> artifacts so users can use these components without rebuilding on their own.
> Additionally, it doesn't make sense for them to be in bin/ since they're not
> user-facing in any way (i.e a user would never directly invoke them). So I
> would propose putting them in an sbin directory.
> The final proposed path is $HADOOP_HOME/sbin/<platform>/{jsvc,task-controller}
> Note this is not an incompatible change since these components were not
> present in any prior apache release.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
