ci: avoid production-length waits in retry fixtures

[proerror77]

Change

The baseline-reader retry fixture spent about a minute waiting for production timers on every CI policy run. Use a test-local Ruby timer shim only for retry cases. Assert the requested 45-second timeout and 1/2-second backoff values. Trigger the real timeout after one second and retain the real process-group cancellation and dead-process assertion. Record backoff calls instead of sleeping.

All existing retry, rejection, pagination, identity and confidentiality cases remain. Production reader code, its 180-second total budget, workflows, selection and required gates are unchanged. The shim is created inside the fixture directory and scoped through RUBYOPT to its read_baseline calls. Ruby's timeout thread retains real timing.

Issue relationship

None

Validation

• Base: 013b2e12ef6113fd19fbaf2179ca652de88cfaae.
• Complete local fixture passed before/after: 84.038 s → 22.251 s. Candidate retry segment: 6 s. This is local evidence, not a cloud latency claim.
• Separate isolated mutations to the actual reader (timeout 45→44, first backoff 1→2, maximum attempts 3→2) all made the fixture fail.
• Shell syntax, ShellCheck and git diff --check passed. Diff review confirms no production or workflow changes.
• Cloud baseline, same Ubuntu image ubuntu-24.04 version 20261012.334.1: previous PR job 194 s total / 179 s policy step; base main job 166 s total / 152 s policy step.
• Candidate, same source da5a897ea39316a84323b06c76f38ce0de933faa and same Ubuntu image: attempt 1 140 s total / 126 s policy step; attempt 2 116 s total / 101 s policy step. Both passed all existing checks. Retry segment: 7 s in attempt 1.
• Compared with the exact base main job (166 s), full-job savings were 26–50 s (16–30%). Against both baseline samples, means were 180 s → 128 s. These are separate hosted runners with identical image versions, not the same VM or a statistical guarantee. Unlike the prior duplicate-test change, both candidate full-job measurements beat both baseline measurements.
• All three required current-head gates passed. The required Monorepo gate was re-run after the second measurement and passed.

Runtime impact and rollback

Test-only timing. No production timeout, retry policy, release behavior, or runtime changes. Existing automatic publication selection remains unchanged. Revert this commit to restore real waiting in the fixture.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

• Configuration used: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: 957958a1-1e1d-4114-a9b7-a24c688e4162

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-16T07:47:10.133750Z	da5a897	PR opened
🔒 Security Review	✅ Completed 2026-10-16T07:47:37.623956Z	da5a897	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.