Skip to content

Conversation

yosefe
Copy link
Contributor

@yosefe yosefe commented Sep 16, 2018

When the underlying SPML requires active progress on target side (e.g RMA emulation over active messages), need to call opal_progress() to avoid a deadlock.
For example, when UCX is using TCP transport.

@yosefe yosefe merged commit 8bd89b8 into open-mpi:master Sep 20, 2018
@yosefe yosefe deleted the topic/shmem-lock-progress branch September 27, 2018 08:30
@gpaulsen
Copy link
Member

In the future, for PRs like this, it would be nice to have a link to an issue with more details such as:

  • How the issue was discovered
  • Under what circumstances users might hit the bug.
  • Does this result in a hang or just poor performance?
  • What components need to be loaded to hit the issue etc.

The reason this is important for PRs like this, is because additional calls to opal_progress can have wide reaching performance issues, and if any performance regression is traced back to this PR, we'd need to understand if this a bit better.

Thanks.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants