I have been able to reproduce it with a much simpler program, that I have added to the below ticket. It does seem like a bug in mpich, unfortunately, though I'm surprised that we didn't encounter this earlier. Perhaps this shows a hole in our test coverage. -- Pavan

On Apr 25, 2017, at 3:57 AM, Sewall, Jason <[email protected]> wrote:

...

From: Latham, Robert J. [mailto:[email protected]] Sent: Monday, April 24, 2017 3:35 PM

...

I can confirm that with NOLOCAL, some processes are stuck in BARRIER while one process is stuck in MPI_Waitsome

Dear Robert,

That barrier is likely a subordinate call inside the top-level Bcast, no? I don't have any Barriers in the code, as I recall.

The Waitsome issue is indeed what I saw with my printf debugging, although it really only clarified the symptom.

...

I hope someone more familiar with the collectives can suggest a fix. For now, I've triaged this in issue #2609

https://github.com/pmodels/mpich/issues/2609

Let me know if there is something else I can do to help here.

Cheers, Jason

_______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

_______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss