Skip to content

ci: move three non-blocking PR-path rows to nightly LAB (Q6, C324) - #5070

Closed
noahgift wants to merge 1 commit into
mainfrom
ci/4955-lab-rows-r2
Closed

noahgift wants to merge 1 commit into
mainfrom
ci/4955-lab-rows-r2

Conversation

@noahgift

Copy link
Copy Markdown
Contributor

Summary

Moves three PR-path rows that could not block a merge to nightly LAB, under the GATE/LAB rule (C324). The rows are the ARM64 GPU quick row, the x86 CUDA unit row and mac-check. None of them is in gate's needs, and the required checks are gate and workspace-test. A quorum ruled DELETE-TO-LAB 3/3 on each row and PROMOTE 0/3.

  • ci.yml: the gpu-touched, GPU and mac-check jobs are removed, and mg-reuse no longer outputs mac. gate's needs are unchanged.
  • gpu-lab-nightly.yml (new): carries the two GPU rows. It runs nightly and on dispatch, on the same runner labels and in each box's concurrency group. It also includes the ci: sov.* fat-job containers outlive cancelled runs (no teardown on cancel) #4948 teardown step (always(), fat_teardown.sh).
  • nightly.yml: mac-check now runs beside build-darwin. publish does not need it.
  • The guards, tests and contract wording are updated to match: ci_mg_reuse wiring rows, the determinism-wait range and ci_gpu_touched_test. pv validate reports the contract valid.

The fourth row, sov.provenance, is not in this change: skipping it on PRs needs a fat_driver.py edit, which the pre-commit complexity hook refuses. It stays tracked on the issue.

Supersedes #5069. That PR went DIRTY after #5061 and #5066 merged, so this is a re-cut of the same commit onto main 83ef430. The only change in the patch is the text of one deleted line in ci/sections.yml: #5066 had routed the removed selector's shallow fetch through fetch_p6.sh. roadmap-aggregate makes no change.

Test plan

Self-tests on the re-cut tree (main 83ef430):

  • ci_mg_reuse: 78/0
  • ci_gpu_touched_test: 30/0
  • ci_gpu_touched --self-test: 19/0
  • fat_driver: 60/60

Also:

  • determinism-wait, guard-steps-isolated, dag, perf-concurrency, runner-labels and guards-are-wired: PASS.
  • pv validate: valid.
  • actionlint: the same 4 info findings as main.

Approval:

  • Sign-off round (C314): 3-0 at e5244dd, re-bound 2-0 at 4a07b1b.
  • Reviews: two non-author reviews, both APPROVE at 4a07b1b.
  • Operator: yes, under C357 3.

Refs #4955

keep-open: #4955 stays open for the fourth row (sov.provenance).

ont-delta: none (CI workflow and guard wiring only; no ontology term added, removed or renamed)

🤖 Generated with Claude Code

The ARM64 GPU quick row, the x86 CUDA unit row and mac-check could not
block a merge: none is in `gate`'s needs, and the required checks are
`gate` and `workspace-test`. Under the GATE/LAB rule a check that cannot
block never runs on a PR, merge-queue or release path. A quorum ruled
DELETE-TO-LAB 3/3 on each, PROMOTE 0/3.

- ci.yml: the gpu-touched, GPU and mac-check jobs are gone; mg-reuse
  no longer outputs `mac`. `gate`'s needs are unchanged.
- gpu-lab-nightly.yml (new): the two GPU rows, nightly + dispatch, on
  the same labels, in each box's concurrency group.
- nightly.yml: mac-check beside build-darwin; publish does not need it.
- Guards and tests follow: ci_mg_reuse wiring rows, determinism-wait
  range, ci_gpu_touched_test, contract wording (pv validate: valid).

The fourth row, sov.provenance, is not in this change: skipping it on
PRs needs a fat_driver.py edit, and that file already exceeds the
pre-commit complexity limits, so the hook refuses any edit to it. It
stays tracked on the issue.

Self-tests: ci_mg_reuse 78/0, ci_gpu_touched_test 30/0, ci_gpu_touched
19/0, determinism-wait 26/0, fat_driver 60/60; dag, perf-concurrency,
runner-labels, guards-wired, guard-steps-isolated PASS; actionlint adds
no finding.

Rebased onto main 9e77499. Main had since added the #4948 teardown
step (always(), fat_teardown.sh) to the three GPU jobs here; the jobs
stay removed, and the step moves to both jobs of gpu-lab-nightly.yml.
Re-run on the rebased tree: ci_mg_reuse 78/0,
ci_gpu_touched_test 30/0, ci_gpu_touched 19/0, fat_driver 60/60;
determinism-wait, guard-steps-isolated, dag, perf-concurrency,
runner-labels, guards-are-wired PASS; pv validate: valid; actionlint
findings on ci.yml and nightly.yml equal main's (4 info).

Re-cut onto main 83ef430. #5066 had routed the PR-diff selector's
shallow fetch through fetch_p6.sh; that selector block is removed here,
so the only change from the reviewed patch is the text of one deleted
line. roadmap-aggregate-check: ok.

Refs #4955

Agent: aprender-wmainred
Agent: apr-4918
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions

Copy link
Copy Markdown

§13.11 rung 1 — quorum shadow verdict

S13-SHADOW pr=5070 head=4a07b1baecce817792db6683e5a4ad066e1965c4 verdict=REFUSE class=Q1 arm_rc=1

Shadow mode: this records a verdict and merges nothing. A refusal
to arm is not a block (§13 adds zero rows to §7) — the pull request is
exactly as green as it was.

@noahgift
noahgift added this pull request to the merge queue Oct 10, 2026
@github-merge-queue
github-merge-queue Bot removed this pull request from the merge queue due to a conflict with the base branch Oct 10, 2026
@noahgift

Copy link
Copy Markdown
Contributor Author

Superseded by #5076: a re-cut of the same commit onto current main, after this PR went DIRTY.

@noahgift noahgift closed this Oct 10, 2026
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