Route trusted Naruon Strix metadata to existing control runners - #2442
seonghobae wants to merge 5 commits into
Conversation
|
Warning Review limit reachedNext included review available in 8 minutes. View limit detailsLimit details: You’ve used the included review currently available. You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Review configuration: ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Advanced Run ID: 📒 Files selected for processing (3)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
…rol-admission-20260927
…rol-admission-20260927
…ission The open head was 33 commits behind. Those commits include the sidecar runtime and coverage-checkout repairs that the current review runs failed on. Naruon stays on the five API-only control jobs; the model scan stays hosted.
|
Exact-head admission correction — Ready is review admission only. Fresh audit against base
This PR is moved to Draft/Proposed until the causal owner repair is present on a successor exact head and re-audited. Queued/pending work is neither an additional blocker nor passing evidence. No Close, force push, destructive rebase, manual rerun, synthetic status/approval, merge, auto-merge, or bypass was performed. |
This branch has not been deployed
1 participant
Latest integration checkpoint
Current tested head:
942b960b8f5744f164bba16d1641d1f8a55b9888. Ordinary merge of central main422defa2f03645b54dccc4d0da6e332a131dae75preserves the independently accepted release-gate changes. The unique three-dot diff remains the Strix Naruon metadata routing, its contract assertion, and RCA documentation.At this head, routing and real continuation-shell tests pass locally (7 in 5.12s) and with GITHUB_ACTIONS=true (7 in 4.46s), with warnings as errors and plugin/capture loading disabled. actionlint and diff checks pass. Current main rules were successfully retrieved before push; hosted checks and independent review are still required.
Latest observation: the three control runners are online and busy. Naruon #1792 Strix admission job 108638506812 is queued requesting ubuntu-24.04 with runner_id=0. This is a later snapshot than the idle-runner observations below; no current idle-capacity claim is made. Its central required-workflow path is absent from the target repository tree, so the target run metadata alone cannot prove the injected workflow source/ref. No cancellation or runner-group mutation was performed.
Historical checkpoint details follow; their test counts and state observations apply only to the heads and times stated.
RCA and change
Naruon current #1789 head 8b7e3d03236073191e4945c03eed18d02471c336 run36320590274 queues metadata jobs on ubuntu-24.04 with runner_id=0, while the existing control group has idle online runners. Today's deployed Strix selector permits .github and fast-mlsirm but excludes Naruon. Hosted Semgrep on #1800 succeeds, so evidence does not establish a global hosted outage or full capacity root cause.
Reuse the existing control allocation for Naruon's changed-scope, current-head admission, superseded-run cleanup and manual evidence publication. Preserve the exact central workflow main ref, dedicated group and labels, no PR checkout, existing permissions and hosted model scan. No provider/model timeout changes, runner-group grant mutations or cancellations.
Verification
New Naruon assertion fails against the unchanged workflow. Runner/scope contracts: 13 passed locally and 13 with GITHUB_ACTIONS=true. actionlint and git diff --check passed. Live group6 visibility all/public allowed and selected central Strix main source verified. No new capacity allocation.
Deployment boundary
Existing runs keep their old workflow source. Native assignment for a newly admitted exact head after protected merge is required; this PR does not claim all checks green, measured latency improvement or full Story1.3 completion.
Admission revalidation — 2026-09-27 14:54 UTC
Native Strix job
108624881622in run36321072667remains queued requesting self-hosted/linux/x64/cwlab-control while group6 reports online idlecwlab-s1-05. No pending deployment approval was returned. This rules out treating every observed wait as simply all control workers being busy; it does not prove the organization concurrency ceiling is the current cause.An earlier group snapshot lacked OpenCode central-main access. Immediate no-cache revalidation returned the already-deployed
opencode-review.yml@refs/heads/mainentry, so the proposed restoration was unnecessary and no group mutation was performed. No lost-update/revocation cause is asserted.REST run-list reads continue to fail on shared quota; a bounded GraphQL sample of recent central-main commit check suites confirms queued work but is not a complete active-run census. The specific live OpenCode job
108600597308remains in sidecar provisioning at sourcee6334e229581a918e2f22de18733b76fa65d7e71; this does not establish model inference activity or authorize duration-only cancellation. Existing merged preflight/continuation repair #2440 remains its source owner; no duplicate repair or runner restart was introduced.Current-main continuation carryover
Exact integration head: 9e3a695. Ordinary merge of main e07c7e1 preserves the independently merged Strix capacity classifier and continuation. Extend Naruon control routing to its new API-only continuation job as well; all five metadata jobs retain the exact trusted workflow-ref guard, hosted fallback, and no PR checkout. The model scan remains hosted.
Final routing and actual continuation-shell checks: 7 passed locally and 7 with GITHUB_ACTIONS=true (plugin autoload disabled, capture disabled); actionlint and diff checks passed. Initial globally loaded pytest runs remain live without result and are not acceptance. A broader warnings-as-errors readiness run separately reported 3 failures from HTTPError cleanup ResourceWarnings on Python 3.14; no production readiness code was changed here. Main rules were retrieved earlier, but the final refresh returned shared REST quota 403; no protected merge claim.
RCA correction: #2440 changes Noema only. Shared readiness starvation is repaired by merged #2411. OpenCode run 36308627125 remains live at old immutable source e6334e2; its sidecar step also performs inference readiness, so duration is not proof of failure and was not used to cancel it.