[codex] Fix translation worker queue progress

[riderx]

Summary

• Bump the translation cache version so production skips bad partial state from the previous worker.
• Let queue jobs translate a full page worth of batches instead of one batch per invocation, while persisting partial state between batches.
• Avoid expensive cache generation from HEAD misses and keep the original string when the model repeatedly drops a protected brand/developer token during single-string fallback.

Root Cause

Production had the latest deploy, but /fr/ and /id/ only had partial R2 state and no completed page object. The queue was flooded by crawler-triggered language URLs, and each page only advanced one batch before being requeued behind more cold work. Logs also showed retries caused by protected-token drops such as Capacitor, which failed whole queue jobs.

Validation

• bun run ci:verify:translation
• bun run verify:real-translation
• bunx wrangler deploy --dry-run

Summary by CodeRabbit

• Bug Fixes

  • Improved translation reliability by preserving original text when certain errors occur.

• Performance

  • Enhanced translation processing efficiency through optimized batching.
  • Updated translation cache system for better performance.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 87e48db2-47dc-4d43-ad37-1f4c9701854b

📥 Commits

Reviewing files that changed from the base of the PR and between 19cdbc2 and 809ec9c.

📒 Files selected for processing (1)

• apps/translation-worker/src/index.ts

---

📝 Walkthrough

Walkthrough

The PR updates the translation worker with a cache version bump to v3, increases batch-per-job capacity from 1 to 96, adjusts batch state management within the translation loop, adds error handling for protected token drops, and adds early termination for non-GET requests.

Changes

Translation Worker Optimization & Error Handling

Layer / File(s)	Summary
Configuration & Constants apps/translation-worker/src/index.ts	Cache version incremented to 2026-05-02-llama-3.1-8b-json-body-v3; translation batches per queue job increased from 1 to 96.
Batch Processing State Management apps/translation-worker/src/index.ts	In refreshCacheIncrementally, partial translation state is now written inside the batching loop when per-job batch limit is reached and more batches remain, enabling correct continuation rather than only writing after loop exit.
Error Handling & Fallback apps/translation-worker/src/index.ts	In translateSingleText, detects "protected token drop" errors; logs and returns original source text instead of proceeding to final failure.
Request Routing apps/translation-worker/src/index.ts	In serveTranslated, adds early guard returning temporary English redirect for non-GET requests before cache lookup.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• [codex] Chunk translation queue jobs #615 — Adjusts batch-per-job limits and moves partial-state writes inside the batching loop to correctly continue incremental chunked translation and partial-state handling introduced in that PR.
• Validate translation worker on real pages #620 — Directly overlaps on changes to TRANSLATION_CACHE_VERSION and batching parameters/batch-handling logic.
• Fix translation worker HTML parsing and HEAD cache refresh #608 — Modifies how translation worker handles non-GET requests; this PR returns a temporary English redirect in serveTranslated while the retrieved PR converts non-GET to GET before rendering.

Poem

A rabbit hops through batches swift, 🐰
From one to ninety-six in shift,
With guarded tokens, states that write,
And GET requests kept shining bright,
The cache hops on, refreshed and tight!

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/fix-translation-prod-queue

---

_{Review rate limit: 1/5 review remaining, refill in 36 minutes and 19 seconds.}

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

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud