Skip to content

[CI] Parity: skip missing configs without aborting report - #3558

Open
ethanwee1 wants to merge 1 commit into
ethanwee/parity-preview-replacementfrom
ethanwee/parity-missing-config-develop
Open

[CI] Parity: skip missing configs without aborting report#3558
ethanwee1 wants to merge 1 commit into
ethanwee/parity-preview-replacementfrom
ethanwee/parity-missing-config-develop

Conversation

@ethanwee1

@ethanwee1 ethanwee1 commented Aug 11, 2026

Copy link
Copy Markdown

Summary

When a per-architecture ROCm config or CUDA Inductor baseline did not run for a SHA, record the gap and continue downloading the remaining configs instead of aborting the whole report. The final nonzero status still exposes incomplete coverage.

Clean replacement for #3446 on current develop, stacked on #3554. Conflict resolution intentionally retained current fallback shard behavior; #3535 remains the dynamic topology dependency.

Test plan

  • Compile download_testlogs
  • Verify no stale nightly, conflict markers, or unresolved discovery helper calls
  • Exercise one missing ROCm config and one missing CUDA Inductor baseline

Made with Cursor

Alignment landing order

Depends on #3554 and #3535. Retarget to develop after those land; land after #3559 and before #3557.

… the run

mi300's default/distributed/inductor come from three separately-scheduled
workflows (rocm-mi300 / periodic-rocm-mi300 / inductor-rocm-mi300) that do not
always land on the same SHA. When one config's workflow run did not run for a
SHA, download_testlogs raised "workflow not found ... increasing max_pages" and
aborted the whole download -- dropping the configs that DID run and producing no
report. (The message was also misleading: download_workflow_run already filters
by head_sha, so it's not a pagination problem -- the run genuinely doesn't
exist for that SHA.)

The parity-auto gate can't prevent this per-arch: it dispatches arches in one
combined run, so its dispatch-level exclude flags can't drop a config for just
one arch. Handle it in download_testlogs (which runs per-arch): when a config's
ROCm workflow (default/distributed/inductor) or the CUDA inductor baseline is
absent for the SHA, log it, record it in error_msgs, and skip just that config.
The report is still built from the available configs, and the job is still
flagged failed at exit so the gap stays visible.
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