PML/UCX: fixed hang on MPI_Finalize - v3.1.x #5248
Conversation
|
@hoopoepg Is this a cherry pick from master? Or is this unique to v3.1? If it's a cherry pick from master, please be sure to use In general, we usually commit fixes to master first and then file PRs that are cherry picks from master to the release branches -- unless there's a fix specific to a release branch (and not relevant to master), of course. But then we specifically note that in the PR so that reviewers aren't surprised. |
hoopoepg
force-pushed
the
topic/pml-ucx-hang-on-finalize-v3.1
branch
from
855d6a2 to
ae1a54d
Compare
fixes issue openucx/ucx#2656 added flush for worker object to complete all pending operations Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com> (cherry picked from commit 0a8261f)
hoopoepg
force-pushed
the
topic/pml-ucx-hang-on-finalize-v3.1
branch
from
ae1a54d to
923e159
Compare
|
@jsquyres yep, sorry, forgot to note about cherry-pick |
|
Ok, thanks. Does it need to go to v3.0.x as well? |
|
I think yes, it seems this is long-play issue |
|
yes |
- added debug output for woker flush failure Signed-off-by: Sergey Oblomov <sergeyo@mellanox.com> (cherry picked from commit c3dbc86)
|
bot:mellanox:retest |
alinask
changed the title
|
@bwbarrett @jsquyres Can you please merge this PR? |
5 participants
fixes issue openucx/ucx#2656
added flush for worker object to complete all pending operations
cherry picked from #5227, commit 0a8261f
Signed-off-by: Sergey Oblomov sergeyo@mellanox.com