WeeklyTelcon_20230523

Open MPI Weekly Telecon ---

• Dialup Info: (Do not post to public mailing list or public wiki)

Attendees (on Web-ex)

• Jeff Squires (CISCO)
• Geoff Paulsen (IBM)
• Howard Pritchard (LANL)
• Luke Robison (Amazon)
• Thomas Naughton (ORNL)
• Tommy Janjusic (nVidia)
• Amir (ORNL)

New Issues:

• Accelerator aware (OFI) NIC selection: https://github.com/open-mpi/ompi/issues/11696
• https://github.com/open-mpi/ompi/issues/11681
  • Fix: https://github.com/open-mpi/ompi/pull/11683 Propagate the error from the generalize request free cb to user

v4.1

• No new updates
• PRs coming in as needed.
• A release to roll out end of summer, unless someone needs something sooner.

v5.0

• PMIX v4.2 async modex issue: https://github.com/openpmix/openpmix/issues/3077

  • Work around: -x PMIX_MCA_gds=hash
  • Need to evaluate performance impact.
  • Tommy will investigate when the issue was introduced and hopefully speed up a potential fix.

• MAC Params issues are biggest issues now - no new updates.

  • https://github.com/openmpi/ompi/issues/11532
  • https://github.com/openpmix/prrte/issues/1731
  • Plan is to have 2 of the 3 fixes for v5.0.0, 3rd issue can wait for 5.0.x
  • Quincy assigned, working on docs first.

• Need to cherry-pick NIC selection (distances PR fixes) to v5.0.x

  • Several PRs will go into main, including coverity fixes.
  • Amir to open up a v5.0.x PR to track all main commits and cherry-pick to v5.0.x when finished.

• UCX and enable mca dso do not mix issue: https://github.com/open-mpi/ompi/issues/11632

  • Fix: https://github.com/open-mpi/ompi/pull/11640
  • Fix is good as is; however, we want to investigate @hjelmn suggestion 'deregistering group as is done in mca_base_component_unload'.
  • If this is requires a lot of work, then we'll merge this PR (since it's like how other places in the code base solve the problem), and defer the real fix to #11680, which doesn't need to block v5.0.0.
  • If what @hjelmn is suggesting is easy, then let's go ahead and fold it into this PR, merge it, and cherry pick to v5.0.x in time for v5.0.0.