fix: redirect cold translation misses

[riderx]

What changed

• Switch the translation worker default model and Wrangler config from Llama to Cloudflare Workers AI Kimi K2.6: @cf/moonshotai/kimi-k2.6.
• Parse OpenAI-style choices[].message.content responses so Kimi output is accepted.
• Change cold-cache localized requests to immediately issue a temporary redirect to the English path while translation generation continues in the background.
• Keep cached localized pages fast: existing cache is served, stale cache is served while refreshing, and only empty cache redirects.
• Bump the worker cache version for the model/runtime behavior change.

Validation

• bun run ci:verify:translation

Summary by CodeRabbit

• New Features

  • Unavailable translations now automatically redirect to the default language version instead of showing error messages, enabling a more seamless user experience across different language preferences and regions.

• Improvements

  • Translation request response times significantly enhanced through asynchronous background cache refresh operations without blocking.
  • AI translation model and cache versioning updated for improved performance.

[coderabbitai]

Caution

Review failed

Pull request was closed or merged during review

📝 Walkthrough

Walkthrough

The translation worker updates its AI model from Llama-3.1-8b to Kimi-K2.6 and replaces synchronous 503 error responses with asynchronous cache refresh behavior. On cache misses, the worker returns a 302 redirect to the default locale while scheduling background translation generation, and expands AI response parsing to handle OpenAI-style choice structures.

Changes

Cohort / File(s)	Summary
Model & Configuration apps/translation-worker/wrangler.jsonc	Updated TRANSLATION_MODEL from @cf/meta/llama-3.1-8b-instruct to @cf/moonshotai/kimi-k2.6.
Worker Logic & Response Handling apps/translation-worker/src/index.ts	Updated DEFAULT_MODEL constant and TRANSLATION_CACHE_VERSION to 2026-05-01-kimi-k2.6-v1. Replaced synchronous 503 error responses with 302 temporary redirects on cache misses. Changed cache refresh to execute asynchronously in background via ctx.waitUntil() instead of awaiting completion. Enhanced extractAiText() to parse OpenAI-style choices array structures. Added null body handling for HEAD requests.

Sequence Diagram

sequenceDiagram
    actor Client
    participant Worker as Translation Worker
    participant Cache
    participant AI as AI Model

    Client->>Worker: Request translation
    Worker->>Cache: Check cache
    alt Cache Hit
        Cache-->>Worker: Translation found
        Worker-->>Client: 200 + Translation
    else Cache Miss
        Cache-->>Worker: Not found
        Worker->>AI: Schedule background refresh
        activate AI
        Worker-->>Client: 302 redirect to default locale
        Note over Worker: Response sent immediately
        AI->>Cache: Generate & store translation
        AI-->>Worker: Done (background)
        deactivate AI
    end

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• Fix translation worker HTML parsing and HEAD cache refresh #608: Modifies the same refreshCache() logic in the translation worker to ensure refresh requests use GET-equivalent semantics, complementing the asynchronous cache refresh pattern introduced here.

Poem

🐰 A rabbit hops through translation streams,
Swapping models—oh, such dreams!
Cache misses dance, redirects flow,
Background refreshes steal the show,
Async magic makes things bright! ✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'fix: redirect cold translation misses' directly reflects the main change: handling cold cache translation requests with redirects instead of 503 errors.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ 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/use-kimi-translation-model

---

_{Review rate limit: 0/5 reviews remaining, refill in 51 minutes and 59 seconds.}

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

[sonarqubecloud]

Quality Gate failed

Failed conditions
C Reliability Rating on New Code (required ≥ A)

See analysis details on SonarQube Cloud

Catch issues before they fail your Quality Gate with our IDE extension SonarQube for IDE

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5a492c5fb1

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Return origin status on cold misses for non-cacheable pages

On a cache miss this branch always returns a 302 to the English URL, but refreshCache only stores responses when response.ok && isHtmlResponse(response). That means localized requests whose English origin returns a redirect, 404, or any non-HTML response will never populate cache and will be permanently redirected to English on every request (e.g. /fr/... pages that should be 301/404). This is a behavior regression from the previous flow, which returned the actual origin status on first request.

Useful? React with 👍 / 👎.