Skip to content

OSC/UCX: Fix early cleanup + race conditions #6754

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 2 commits into from
Jun 22, 2019
Merged

OSC/UCX: Fix early cleanup + race conditions #6754

merged 2 commits into from
Jun 22, 2019

Conversation

@artpol84
Copy link
Contributor

  • Fix early initialization of the worker pool (WP). WP has to remain until the finalization of the OSC component.
  • Ensure the atomicity of the OSC/ucx global state updates.

@artpol84 artpol84 added this to the v5.0.0 milestone Jun 11, 2019
@artpol84 artpol84 requested review from janjust and xinzhao3 June 11, 2019 21:36
@artpol84 artpol84 self-assigned this Jun 11, 2019
janjust
janjust reviewed Jun 12, 2019
View reviewed changes
janjust
janjust reviewed Jun 12, 2019
View reviewed changes
janjust
janjust approved these changes Jun 12, 2019
View reviewed changes
Copy link
Contributor

@janjust janjust left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@artpol84 I assume this was tested, if so this looks good to me.

@open-mpi open-mpi deleted a comment from ibm-ompi Jun 12, 2019
@open-mpi open-mpi deleted a comment from ibm-ompi Jun 12, 2019
@jjhursey
Copy link
Member

bot:ibm:retest

artpol84
artpol84 commented Jun 12, 2019
View reviewed changes
@MrBr-github
Copy link

bot:retest

derbeyn
derbeyn reviewed Jun 20, 2019
View reviewed changes
artpol84
artpol84 commented Jun 20, 2019
View reviewed changes
artpol84 added 2 commits June 20, 2019 09:05
@artpol84
osc/ucx: Fix worker pool finalization
0857742 
Signed-off-by: Artem Polyakov <[email protected]>
@artpol84
osc/ucx: Fix possible win creation/destruction race condition
6678ac0 
To avoid fully initializing the osc/ucx component for MPI application
that are not using One-Sided functionality, the initialization happens
at the first MPI window creation.

This commit ensures atomicity of global state modifications.

Signed-off-by: Artem Polyakov <[email protected]>
@artpol84
Copy link
Contributor Author

artpol84 commented Jun 20, 2019 •
edited
Loading

@derbeyn Thanks for the review.
I've only runtime tested 0857742 and this works well.
6678ac0 needs testing still.
@janjust is working on it. During the testing, I believe we would catch that lock=unlock bug, but thanks for catching it early.

@artpol84
Copy link
Contributor Author

We runtime confirmed that 6678ac0 does fixes the race condition.
The following test was used:
https://github.com/artpol84/poc/blob/master/MPI/MPI_one-sided/win_test.c

@artpol84 artpol84 merged commit 6433da7 into open-mpi:master Jun 22, 2019
@janjust janjust mentioned this pull request Aug 12, 2019
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@janjust janjust janjust approved these changes

@xinzhao3 xinzhao3 Awaiting requested review from xinzhao3

+1 more reviewer

@derbeyn derbeyn derbeyn left review comments

Reviewers whose approvals may not affect merge requirements

Assignees

@artpol84 artpol84

Labels

bug Target: v5.0.x

Projects

None yet

Milestone

v5.0.0

Development

Successfully merging this pull request may close these issues.

5 participants

@artpol84 @jjhursey @MrBr-github @janjust @derbeyn