Meeting 2019 11 07 - openpmix/openpmix GitHub Wiki

Assorted Meeting Notes:

  • 'make check'
    • Part of the problem was in the test themselves (some reporting success even though they fail)
      • Unit tests needed to be repaired (across all of the release branches)
      • Need to make sure unit tests use the limited capabilities of a specific release branch.
      • Some of the tests we shipped in the 2.x series don't pass even within the same release
      • Removed event handler code from the tests. Existing version could throw the program into an infinite loop.
      • Disabled multi-server functionality. Force everything to be a single server.
        • Technique for doing it was not quite correct.
          • nodemap was only the local procmap instead of the global procmap
      • tests/simple = are ok tests = had issues
    • Disabled 'make check' for all but 2.2, 3.1, and master
      • Other branches (2.0 lost cause) have unit tests that are far too old
        • 2.1 past most things, but not everything.
        • 2.0 failed most/all of the test
        • 1.2 too old
    • A few minor changes to the library outside of the test suite:
      • nothing to major.
      • test_resolve_peers was having the most trouble
    • To do:
      • re-enable the multi-server functionality
      • if folks are so interested they can go back and fix the old release branches (2.1, 2.0, 1.2)
  • Should we centralize the 'tests' directory into pmix-tests?
    • Some tests are subtlety different in different release versions (as well as behaviors)
      • we would need some configurey logic to help with that discovery
    • This should make maintaining the tests a bit easier since they are all grouped together.
    • Still ship a snapshot of this directory with each release.
    • (Josh) action item to pursue this
  • Should we do a 3.2?
    • Timing wise 4.0 will come out soon, so should we wait?
    • We would need to re-branch from 3.1
      • Then port the dstore optimizations to the new 3.2 branch
    • Feeling is that it is worth doing.
    • (Ralph) will re-fork v3.2 and make a first attempt at bringing the dstore optimizations
  • 3.1.5 and 3.2 and master
    • (Mellanox/IBM) should re-verify the PMIx_Get performance given new conditions before the fastpath
  • GitHub Actions - Mellanox CI
    • https://github.com/openpmix/openpmix/pull/1514
    • Ralph enabled the GitHub Actions on the openpmix repo
    • Tests may be failing because of the multi-server 'make check' issue
    • Where will we see these testing results?
      • Actions seems to be firing, but not showing up in the CI box with the PR
      • (Artem) to follow up with their dev ops team, and send a note to the mailing list
  • SC'19 Prep
    • Wed, 12:15- 1:15 in 210-212
    • Will discuss both OpenPMIx, PRRTE, and PMIx Standard effort
    • Anything we should highlight: Community Highlights
    • Please send slide updates by Nov. 15
  • Developer Deep Dives
    • Topics:
      • event handling (Ralph)
      • GDS and dstore (Artem)
      • cross-version handling (Ralph)
    • Structure
      • High level design/architecture
      • Guidance on how the parts connect to each other
        • And maybe where the parts are located in the source tree)
      • Then drop into the code to show where the parts connect
    • Josh will coordinate with speakers on a good date/time for the meeting
  • v4.0 release - looking at the end of the year
    • Gating:
      • Tools documentation for the v4.0 standard document
      • Finishing up the Python interface (store_internal may be last part)
        • Maybe mark it as a preliminary release for broader exposure and early testing
  • OpenPMIx master
  • PRRTE: