You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
.github queued runs hit the API cap of 1,000. By type: CodeQL Scan Dispatch 650, OpenCode Review Dispatch 163, Required Noema Review 35, others static.
Queued OpenCode dispatches for the same PR at different heads: 0. Cancelling superseded runs would not reduce the queue.
Of the 30 most recently completed OpenCode Review Dispatch runs: 22 cancelled, 8 failed, 0 succeeded.
The failures are all intended validate-pr-metadata rejections (head changed or PR closed). Example: run 36603201268.
The cancellations were replaced by a newer head after waiting about 3 hours. Examples: 36620067496 (fast-mlsirm#2273), 36623143994, 36620314334.
In opencode-review-dispatch.yml, all three jobs (validate-pr-metadata, coverage-evidence and opencode-review-target) use runs-on: group: CWL central OpenCode.
Runner group CWL central OpenCode (id 5) contains only cwlab-s1-04.
coverage-evidence has timeout-minutes: 300. A single dispatch can hold that one runner for up to 5 hours, so throughput is at most one PR at a time.
Meanwhile, the CWL central control group (id 6: s1-01, s1-05, s2-01, s1-07) has an idle runner, cwlab-s1-07 (busy=false). That group's selected_workflows does not include opencode-review-dispatch.yml, so it can't take the work.
Evidence commands:
gh api orgs/ContextualWisdomLab/actions/runner-groups/5/runners
Options (operator decision: this changes a security-isolation boundary, so I have not changed it myself)
A. Add more runners to the CWL central OpenCode group, for example moving cwlab-s1-07 from group 6. This keeps isolation per group and widens throughput.
B. Add opencode-review-dispatch.yml@refs/heads/main to group 6's selected_workflows. The isolation trade-off needs review.
C. Short term: run scripts/ci/opencode_queue_priority.py (operator runbook) to put maintainer-labelled review-priority PRs first.
Side note
The defect in #2273 (coverage build context missing requirements-noema-document-ci-hashes.txt) appears fixed in the current main opencode-review-dispatch.yml L718, which now stages that file. Please check whether #2273 can be closed.
Symptom (2026-09-30 08:30 KST / 2026-09-29 23:30 UTC)
.githubqueued runs hit the API cap of 1,000. By type: CodeQL Scan Dispatch 650, OpenCode Review Dispatch 163, Required Noema Review 35, others static.validate-pr-metadatarejections (head changed or PR closed). Example: run36603201268.36620067496(fast-mlsirm#2273),36623143994,36620314334.Root cause
opencode-review-dispatch.yml, all three jobs (validate-pr-metadata,coverage-evidenceandopencode-review-target) useruns-on: group: CWL central OpenCode.CWL central OpenCode(id 5) contains onlycwlab-s1-04.coverage-evidencehastimeout-minutes: 300. A single dispatch can hold that one runner for up to 5 hours, so throughput is at most one PR at a time.CWL central controlgroup (id 6: s1-01, s1-05, s2-01, s1-07) has an idle runner,cwlab-s1-07(busy=false). That group'sselected_workflowsdoes not includeopencode-review-dispatch.yml, so it can't take the work.Evidence commands:
gh api orgs/ContextualWisdomLab/actions/runner-groups/5/runners.../runner-groups/6selected_workflowsgh api orgs/ContextualWisdomLab/actions/runners.github/workflows/opencode-review-dispatch.ymlL56/L399/L2514 (runs-on), L402 (timeout-minutes: 300)Options (operator decision: this changes a security-isolation boundary, so I have not changed it myself)
CWL central OpenCodegroup, for example movingcwlab-s1-07from group 6. This keeps isolation per group and widens throughput.opencode-review-dispatch.yml@refs/heads/mainto group 6'sselected_workflows. The isolation trade-off needs review.scripts/ci/opencode_queue_priority.py(operator runbook) to put maintainer-labelledreview-priorityPRs first.Side note
The defect in #2273 (coverage build context missing
requirements-noema-document-ci-hashes.txt) appears fixed in the current mainopencode-review-dispatch.ymlL718, which now stages that file. Please check whether #2273 can be closed.🤖 Generated with Claude Code