env update: defer hardware smoke to PR review; fix gh repo + manifest note - #1
Merged
Conversation
… note The weekly env-update cron runs on the jumpbox (Alpine musl, no GPU). Its smoke step (`uv run --project envs/<p> python envs/smoke.py --gpu`, run locally — `forge run` only leases the remote host) could never pass there: the CUDA envs can't even build (bitsandbytes/torch ship glibc-only wheels) and there is no GPU to matmul on. Every run recorded ok:false on a smoke failure that was environmental, not a real regression. Since the PR is never auto-merged, defer hardware smoke to the reviewer: - cmd_update no longer runs smoke; marks it deferred with the smoke_host. - report `ok` now means "all locks re-resolved without error" — exactly what the jumpbox can attest. - _pr_body embeds the exact per-profile smoke command to run on the smoke_host before merging. - gh pr create gets `-R GoodAncestor/forge` (origin uses a github-forge SSH alias gh can't map to a repo on its own) — the other half of the pr_error. - drop the now-meaningless --no-smoke flag. Also: correct the stale gutenberg-mirror manifest note (~60GB -> >150GB and growing; size_bytes stays null as an unmeasurable rsync tree).
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The weekly env-update cron runs on the jumpbox (Alpine musl, no GPU). Its smoke step could never pass there: the CUDA envs cannot build (bitsandbytes/torch ship glibc-only wheels) and there is no GPU to matmul on.
forge runalso executes the recipe locally (the--hostonly leases the remote node), so--gpunever had hardware. Every weekly run recordedok:falseon an environmental smoke failure, not a real regression.What changed
Since the PR is never auto-merged, hardware smoke is deferred to the reviewer:
cmd_updateno longer runs smoke; it marks smokedeferredwith thesmoke_host.oknow means all locks re-resolved without error — exactly what the jumpbox can attest._pr_bodyembeds the exact per-profile smoke command to run on thesmoke_hostbefore merging.gh pr creategets-R GoodAncestor/forge(origin uses agithub-forgeSSH alias gh cannot map to a repo on its own) — the other half of the oldpr_error.--no-smokeflag.Also corrects the stale
gutenberg-mirrormanifest note (~60GB → >150GB and growing;size_bytesstays null as an unmeasurable rsync tree).Before merging
Run the smoke command from the PR body on each changed profile’s
smoke_host(e.g. ai02 for cuda-ada, alien03 for cuda-blackwell).Tests
pytest tests/test_envupdate.py(8 passed) + env suite (14 passed), ruff clean.