Repository navigation
Release CI: one job, one reserved runner — a release PR must not fan out into 20 queued jobs #4433
Description
OPERATOR, verbatim (2026-09-25, cop session): "the release has 11 jobs??? wtf, merge into one"
Measured on 0.69.5 (#4318, run 36148145139): the release PR's CI fanned out into 20 jobs, and at 14:46Z 11 of them were still queued behind other PRs' jobs on the shared clean-room pool. A release waits on the slowest queue position among 20.
Want: when the head is a release/rc car, CI runs as ONE job on a runner reserved for releases (label release, never taken by non-release jobs). That job runs the same checks sequentially, gate + workspace-test included, and reports the required contexts (ci / gate, workspace-test).
done_when:
- A release-car PR shows 1 job, which starts in ≤60 s while the pool is saturated.
- The same checks are executed: the release job runs every step the fan-out runs, with Σ steps equal and a case table in the PR.
- A non-release PR is unchanged.
- The mutation proof: remove a step from the release job → the parity check goes RED.
kind: code
Folded sub-tickets (TRIAGE-100, 2026-10-14)
These open sub-tickets were folded here as checklist rows and closed as not planned so the open count stays at or under 100. Reopen one when its work resumes; an open PR for it reopens it.
- A merge group whose guard-tree is RED still burns ~2 h of intel workspace-test time before gate reports it #3177 — A merge group whose guard-tree is RED still burns ~2 h of intel workspace-test time before gate reports it
- Lever 0: guard-cargo/guard-tree stop at the first red step — run every guard and report every failure in one pass (RC #4318: 1 failure per run, 15–71 steps dark) #4415 — Lever 0: guard-cargo/guard-tree stop at the first red step — run every guard and report every failure in one pass (RC batch(0.69.5): EPIC #4249 patch train + #4313 f16 prefill — 0.69.5 RC #4318: 1 failure per run, 15–71 steps dark) (branch:
cb/4415-bash-port,PMAT-4415-falsifier-plant) - release: autopilot cascade step and spec T-4 name different publish executors #4880 — release: autopilot cascade step and spec T-4 name different publish executors
- release autopilot: one unattended pass from bump to live on main #4950 — release autopilot: one unattended pass from bump to live on main
- GEN-001 cond. 1: the release step runs make census (and readme-sync) on release/X.Y.Z, with a receipt #4951 — GEN-001 cond. 1: the release step runs make census (and readme-sync) on release/X.Y.Z, with a receipt
- LAB: release_ready.sh not_measured on main: autopilot_found cannot parse ap_policy_applies' subshell (autopilot.sh:76) #4968 — LAB: release_ready.sh not_measured on main: autopilot_found cannot parse ap_policy_applies' subshell (autopilot.sh:76)
Activity
ci(y2): apply the 3-shard preload only when SHARDS=3 (Refs #4424, #4433)
noahgift commentedon Sep 25, 2026
Re-scope (cop, relaying the operator's ruling of 2026-09-25 16:58 Madrid, verbatim):
done_when now covers items 1 and 3: 20 → 5 jobs, with Σ executed equal to today's (case table); the rc tag is cut on a queue-green main sha within 5 min; promotion is gated by receipts on the tag. #4318 finishes as-is.
noahgift commentedon Sep 25, 2026
Item 3 of the operator's 'make today's rulings permanent' (done = merged + enforced): the #4433 ci.yml PR must also ship a Σ-executed invariant check. The check fails when the set of tests executed across the 5 jobs differs from today's set (by test id, not by count), and a mutation that drops one test turns it RED. [U] until merged.
ci(#4433): fat driver — per-section checkout depth, secrets, section …
ci(#4433): collapse 20 CI jobs into 5 fat jobs + 3 verdict jobs
test(ci): fat_driver self-test — 23-row case table (expressions, AnyS…
fix(ci): restore GATE-MUTANTS-RULE markers around the gate's mutants …
fix(ci): artifact download follows the blob redirect without Authoriz…
fix(ci): sov inputs evaluated in the caller's context; section clones…
docs(release): rc = tag on queue-green main sha; cop rulings register (…
fix(ci): vendored sovereign-ci carries its sccache mount as one decla…
roadmap: PMAT-4433 row (5 fat CI jobs + Σ-executed invariant)
60 remaining items
noahgift commentedon Oct 14, 2026
Folded into #5306 as a checklist row (TRIAGE-100, 2026-10-14). Under the cop's ruling, tickets with a P0 label or the operator's own words that are sub-rows of an existing master are folded, with the operator's text quoted on the master exactly as written. Priority and owner carry over on the row.
Owner aprender-78: if you object, say so here or in the inbox and this ticket is reopened. Work on it continues as before; a PR for it reopens it.
Agent: aprender-clerk
ci(shards): list the lib universe on lib part 1; --release fragments …