btl/openib: delay UCX warning to add_procs() #6152
Conversation
|
@hjelmn please review when you have a chance |
|
@hppritcha It looks ok to me on its own. Just waiting on #6184 and it's associated v4.0.x PR before approving. |
|
@ggouaillardet #6184 is moot now, is this still a WIP-DNM situation? |
|
@ggouaillardet has marked this as a blocker on the corresponding master PR #6184, but we just closed that PR without merging because BTL openib has been removed from master. @ggouaillardet describes the seriousness here: #6184 (comment) |
|
@ggouaillardet While this is no longer important for master, it'd be nice for v4.0.x Also is there someone else who could review this? hjlemn may not have the time anymore. |
|
@gpaulsen This PR is currently half baked. At the very least, I have to backport the two commits from #6184 (that was never merged since According to @hoopoepg the outcome is still worst in some cases (a workaround is to disable I will backport the two commits from now and test it on my cluster (I have a single IB QDR port, and no issue with this PR on master). Can you please bring the topic at today's telcon and decide how to move forward ? |
If UCX is available, then pml/ucx will be used instead of pml/ob1 + btl/openib, so there is no need to warn about btl/openib not supporting Infiniband. Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp> (cherry picked from commit open-mpi/ompi@0a2ce58)
…led. Fixes an issue introduced in open-mpi/ompi@0a2ce58 This is a one-off commit for the v4.0.x branch since btl/openib has been removed from master. Refs. open-mpi#6137 Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
Many thanks to Sergey Oblomov for reporting this issue and the countless traces provided when troubleshooting it. This is a one-off commit for the v4.0.x branch since btl/openib has been removed from master. Refs. open-mpi#6137 Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
ggouaillardet
force-pushed
the
topic/v4.0.x/ucx_warning
branch
from
99d8576 to
8da4605
Compare
|
@gpaulsen I backported the other commits and it is running fine in my environment. |
|
Thanks. We discussed in today's Webex and would like to hear from either @xinzhao3, @jladd-mlnx for a suggestion if this should go into v4.0.x. |
|
From a design standpoint, this is a good PR. However, I don't think I'm the right person to review a change like this in the OpenIB BTL. |
|
@hppritcha @ggouaillardet @jsquyres @hjelmn, So, based on what I read, I THINK this PR is ready to go into v4.0.x. Do you agree? Can we remove the WIP label and merge? |
|
@gpaulsen Someone has to review it. |
|
@gpaulsen do you want to merge this PR? I think its ready. |
|
Can one of the admins verify this patch? |
If UCX is available, then pml/ucx will be used instead of
pml/ob1 + btl/openib, so there is no need to warn about
btl/openib not supporting Infiniband.
Signed-off-by: Gilles Gouaillardet gilles@rist.or.jp
(cherry picked from commit 0a2ce58)