[libfabric-users] Progress question

[Biddiscombe, John A.]

Apologies - I'm now doubting my own problem. Even with the extra polling in place, I still do not see rank 0 receive the message.


I will investigate further.


JB

________________________________
From: Libfabric-users <libfabric-users-bounces at lists.openfabrics.org> on behalf of Biddiscombe, John A. <biddisco at cscs.ch>
Sent: 08 April 2020 19:55:49
To: libfabric-users at lists.openfabrics.org
Subject: [libfabric-users] Progress question


Dear List,


I have a problem that can be summarized by saying that just before it exits a main loop, rank 1 does a send to rank 0 (actually an fi_inject), and then never needs to check for messages again. Rank 1 is finished (but it doesn't shut down yet as there are other thigs to do that don't require messaging). Rank 0 never receives the message that rank 1 send. If I add an extra short loop for a fraction of a second and tell rank 1 to poll completions after the fi_inject, then rank 0 does get the message.


Since fi_inject returns 0 indicating success - I assumed all was ok. If I knew always which send was going to be the last, then I could add a short poll in the right place - unfortunately I don't always know what the last send/inject will be.


Is there a right way to handle this (I guess I need a flush operation of some kind, but I'm not sure where to put it).


Thanks


JB


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/libfabric-users/attachments/20200408/828990f2/attachment.htm>