Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [ptp-user] Job still in state "running" while already finished

Yes, please open a bug.

There are a few possibilities for the cause of this problem. The proxy was never designed to be thread safe, so enabling threading in the ORTE library may be resulting in locking and/or race conditions (there is no guidance on calling ORTE library functions from a non- threaded application.) If I can repeat what you're seeing and this is the problem, then I'll have a chance of tracking it down. The other possibility is more problematic. Our proxy calls ORTE library functions that are not used by anyone else, including ORTE. It's possible that one of these has a bug that is only apparent when threading is enabled. Job termination is one that springs to mind. If this is the case, then getting it fixed may be difficult.

Regards,

Greg

On Aug 15, 2007, at 6:46 PM, Mateusz Pabis wrote:

Great, but don't bother!

I've spent a lot of time to figure this out (well, it takes 40minutes to
recompile OpenMPI on my machine, and today I did it few times).
I rolled back to 1.2.2, but behaved just as 1.2.3
I noticed that there is orted process in the background which takes 100% processor time. I've spent some time googling about this, and finally I
found a workaround.

I've compiled OpenMPI also with --with-threads=posix and
--with-enable-mpi-threads and this was the source of all my troubles.
But the solution does not satisfy me - I need threads in my applications.

Should I report a bug into bugzilla about this issue?
It looks like thread lock or sth.

I'm afraid problem is in PTP because running jobs without PTP proxy goes
smoothly.
I'm looking forward to see this problem solved.

Greg Watson wrote:
Hmmm. It's possible that something has been changed in OpenMPI 1.2.3
that has broken this. The latest I've tried is 1.2.2. I'll update to
1.2.3 and see if I can repeat the problem. It would be nice to get the
fix into the 1.1.1 bugfix version of PTP if possible.

Greg

On Aug 12, 2007, at 3:54 PM, Mateusz Pabis wrote:

hi *,

I'm trying to set up PTP in Eclipse. When I run small hello world
application Job1 is still running - even when all processes have already
finished.

[cut]


_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user




Back to the top