Skip to content

fix: clear stale PMI metadata before YAC remaps - #196

Merged
rajeeja merged 1 commit into
mainfrom
fix/yac-pmi-fd
Oct 29, 2026
Merged

rajeeja merged 1 commit into
mainfrom
fix/yac-pmi-fd

Conversation

@rajeeja

@rajeeja rajeeja commented Oct 29, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • clear stale PMI_FD/PMI_RANK/PMI_SIZE before YAC-backed worker remaps
  • apply the same environment to the remote YAC smoke probe
  • replace the misleading MPI-launcher recommendation with the measured singleton diagnosis
  • update Chrysalis/UCAR endpoint scripts and UCAR documentation
  • add regression tests covering all three serialized worker remap paths

Validation

  • 1567 passed, 1 skipped, 44 deselected, 1 xfailed
  • docs build with -W clean
  • shell syntax and shellcheck clean
  • Chrysalis live conservative remap: 10/10 matrix after the fix
  • UCAR live check_remote_yac: conservative_ok: true; full matrix remains 13/14 because the endpoint's remap_to_rectilinear path is still failing independently
  • direct Chrysalis in-worker test: conservative YAC remap succeeds with PMI_FD removed, matching the local result

The fix intentionally does not add an MPI launcher. Testing showed that multi-rank invocation without explicit gather logic returns silently incorrect, non-deterministic results.

@rajeeja
rajeeja merged commit ea34d28 into main Oct 29, 2026
22 of 23 checks passed
@rajeeja
rajeeja deleted the fix/yac-pmi-fd branch October 29, 2026 21:30
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant