I'm afraid it isn't quite that simple, Jeff. We also have the race condition at
startup - multiple procs on the same machine, from the same job, will be trying
to create the session directory tree. At the moment, we see the fact that some
other proc created it and simply create our own entry underneath as required.
So I don't know how to tell the difference between "some other proc from my job
created it first" vs "this is a stale directory and should be deleted".
However, I might be able to rig something up when the daemons start, and for
singletons. Will give that a try
On Feb 4, 2014, at 6:11 AM, Jeff Squyres (jsquyres) <jsquy...@cisco.com> wrote:
> On Feb 3, 2014, at 6:44 PM, Ralph Castain <r...@open-mpi.org> wrote:
>
>>> If I may suggest to test the behavior of 1.7.x... what about this: Have a
>>> test case that creates a bunch of files (from 0 to 65536) in
>>> /tmp/openmpi-sessions-${USER}... before launching an executable without
>>> mpirun... >:)
>>
>> Ick - it will actually only conflict if/when the pid's wrap, so it's a
>> pretty rare issue.
>
>
> Ralph: what do you think about modifying this for 1.7.5? I.e., if the pid
> dir already exists in the session directory, remove it. This is always safe
> to do (assuming /tmp is a local filesystem) because the OS will never use the
> same PID for 2 concurrent processes.
>
> --
> Jeff Squyres
> jsquy...@cisco.com
> For corporate legal information go to:
> http://www.cisco.com/web/about/doing_business/legal/cri/
>
> _______________________________________________
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users
