[JacobStephens2]

Summary

A hung Actions job inherited GitHub's six-hour default and held a Ticket run in CI watch. Every CI job and every PR-reachable Release job now has an explicit execution budget, sized from recent successful runs.

 CI / PR-reachable Release job
-  timeout-minutes: 360 (inherited)
+  timeout-minutes: 5–60 (per-job, measured)

Release's generated limits live in scripts/generate-release.py; the smoke reusable workflow sets limits on its concrete jobs because GitHub does not support timeout-minutes on reusable-workflow calls. Tag-only host, announce, custom-publish-crates and custom-release-notes cannot hold a PR's CI watch, so their timeouts are outside this change.

Closes #1496

Evidence

• Before: python3 -m unittest discover -s scripts -p test_workflow_timeouts.py failed all seven CI-job subtests with inherits GitHub's 360-minute default. Adding Release coverage then failed for all six PR-reachable concrete Release/smoke jobs. No compilation or setup failure was counted as reproduction.
  After: the same command passes both tests, covering all 13 jobs. The confirmed cause was missing job-level timeouts; step-level bounds do not substitute for this contract.
• Generator: python3 -m unittest discover -s scripts -p test_generate_release.py failed before the generator emitted the new execution bounds and passes afterward. The pinned real cargo-dist 0.33.0 regenerated Release; python3 scripts/generate-release.py --check passes.
• Tooling: python3 -m unittest discover -s scripts -p 'test_*.py': 18 passed.
• Affected checks: NEXTEST_TEST_THREADS=3 RUST_TEST_THREADS=3 /usr/bin/time -p scripts/affected-tests main passed. The selector used the full Rust suite via its cargo test fallback (nextest unavailable locally), followed by all three Release-generator tests. Wall time: 1919.02 seconds (31m59s), compilation included. All 80 Rust result summaries were successful. Three test threads follow the repository's guidance for deadline-sensitive tests on busy hosts; hosted CI independently passed nextest on the same head.
• PR CI: CI run 40615900874 and Release run 40615900906 both completed successfully on 649c0ec7. All 18 checks finished (13 passed, five expected skips). The Linux Rust job took 971 seconds of its 60-minute budget; Console 127/900, site 82/600, macOS gate 211/1200, release builds 176–254/1200, Rocky 12/300 and WSL 50/600 seconds. PR policy uses the macOS gate; the full shards are skipped as expected.

Observed runtime and chosen limits

Measured on 2026-10-28 UTC with gh run list --workflow <workflow> --status success (PR, push and schedule events separately) and gh run view <id> --json jobs. Figures are successful jobs' completedAt - startedAt, including setup/cleanup and excluding queue time and skipped jobs. Full run links and rationale are in the runtime record.

Workflow / job	Recent successful execution (seconds)	Timeout (minutes)
CI / console-test	98–182	15
CI / site-test	59–102	10
CI / test, including nightly cargo test	725–983	60
CI / macos-plan	3–4	5
CI / test-macos-gate	232–265	20
CI / test-macos-shard, all four shards	627–934	60
CI / test-macos	3–4	5
Release / plan	19–26	5
Release / build-local-artifacts, Linux and macOS	180–319	20
Release / build-global-artifacts	20–30	5
Release / custom-release-smoke / gate	3–6	5
Release / custom-release-smoke / rocky	11–16	5
Release / custom-release-smoke / wsl	62–66	10

The CI sample comprises three PRs, three main pushes and one nightly run; Release comprises four PRs and three tag runs. Rocky/WSL execution figures use the tag samples because those PR samples skipped their smoke tests. The same reusable jobs can execute on pipeline-changing PRs, including this one. Full suites get at least 3.6 times the slowest observed runtime, browser/gate jobs at least 4.5 times, and Release builds at least 3.7 times. Five-minute orchestration floors and ten minutes for WSL leave room for setup and transfers. These budgets need remeasurement as job workloads grow.

Test seams

• The job-level GitHub Actions configuration in CI and PR-reachable Release jobs.
• The Release generator CLI with an external cargo-dist stand-in and the pinned real generator.

Review coverage

Independent Standards and Spec reviews against main found 0 Standards / 0 Spec findings. Each reviewer read all seven changed files, including generated Release hunks. Unread changed files: none. Generated-output exclusions: none. Reports and each axis's final files-read list are retained in the Session review artifacts (.thirdshift-review-zTkRVU/standards.md, spec.md, summary.json).

Unaddressed findings

• Standards: none.
• Spec: none.
• Security: none.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Revert the workflow limits and generator policy to restore previous behavior.

Blast Radius: CI

A healthy job slower than its budget will end early. Limits include several times the observed execution time; Rust suites retain 60 minutes. This changes job execution bounds and leaves CI watch's behavior intact.

Built with codex codex-cli 0.164.0 · gpt-6.1-sol · xhigh