ci(gpu) + test(pi07): drop temp push trigger + scope CUDA drain (cleanup of #285)#286
Merged
Merged
Conversation
Cleanup left by #285's squash-merge: - Remove the gpu_test.yml `push` trigger I added to dispatch GPU CI on the fix branch — the run on dd42b69 already confirmed the fix. - Address the claude[bot] nit on #285: scope the pre-init CUDA drain (set_device + synchronize + empty_cache) to the path where we own the init. When a prior test had already initialized the PG, NCCL is already bound to that device and re-pinning here can't unbind it, so the drain is only meaningful when we own init.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this does
Cleanup left by #285's squash-merge.
#285 went through with the temporary
gpu_test.ymlpushtrigger I had added on the fix branch to manually dispatch GPU CI without API auth. The maintainer marked the PR ready and squash-merged it before my follow-up cleanup commit landed, so the trigger leaked intomain. This PR drops it.While I'm here, also addressing the
claude[bot]nit from the original review (PR #285):tests/policies/test_pi07_low_level.py— the pre-init CUDA drain (set_device(0)/synchronize()/empty_cache()) was at function scope, so it ran even when an earlier test in the session had already initialized the PG. In that case NCCL is already bound to whichever device that test picked and re-pinning here can't unbind it, so the drain only does useful work on the path where we own the init. Move it inside theif not already_initialized:block alongside the actualinit_process_groupcall. Comment is updated to spell out the asymmetry.No functional change to GPU behavior — the GPU CI run on
dd42b69(Run Pytest on GPUjob 74911614535, 13min, success) already confirmed the underlying fix works.How it was tested
pre-commit run --files tests/policies/test_pi07_low_level.py .github/workflows/gpu_test.yml— all hooks pass.pytest -m "not gpu" -n auto tests/policies/test_pi07_cpu.py tests/policies/test_pi07_low_level.py— 69 passed, no regressions.gpu_test.ymlchange is trigger-only; the gpu-test job body is unchanged so a re-run on the nightly cron (or manualworkflow_dispatch) still exercises the same fix that already passed ondd42b69.How to checkout & try? (for the reviewer)
git fetch origin claude/fix-gpu-tests-TStWF git checkout claude/fix-gpu-tests-TStWF git diff main -- .github/workflows/gpu_test.yml tests/policies/test_pi07_low_level.py pytest -m "not gpu" -n auto tests/policies/test_pi07_cpu.py tests/policies/test_pi07_low_level.pyChecklist
Generated by Claude Code