fix: raise maxDuration to 300s, extend sitemap JS timeout to 240s

[Flotapponnier]

Root cause

Vercel logs confirm Node.js process exited with signal: 9 (SIGKILL) on every /sitemap.xml request since ~Aug 8.

The sequence:

1. unstable_cache in loadAggregateFromBlob can't store the 4.25 MB blob (Next.js Data Cache hard cap is 2 MB), confirmed by repeated log warnings: items over 2MB can not be cached (4252793 bytes)
2. So every sitemap request re-fetches from /api/aggregate (no caching possible)
3. /api/aggregate fetches from the Paris VPS — cold fetches take 18-65 s (FETCH_TIMEOUT_MS = 65_000)
4. maxDuration = 60 → Vercel sends SIGKILL at 60 s before the fetch can resolve
5. SIGKILL discards the buffered response → Vercel returns 500

The prior JS timeout at 55 s was designed to fire before SIGKILL and return the static fallback, but the background buildFullSitemap() promise (holding the open 65 s fetch) kept the event loop alive past 60 s, preventing clean exit.

Fix

• maxDuration 60 → 300: gives the function time to either complete the full build or exit cleanly after the JS timeout fires
• JS timeout 55 s → 240 s: allows the aggregate fetch (≤65 s) + HL-builder Prometheus sweeps to complete before falling back to the static sitemap

What to do after merging

The workflow file (.github/workflows/prod-deploy.yml) also needs /api/aggregate added to the warm-up step so the ISR cache is hot before the smoke test. That requires workflow scope on the push token — do it as a follow-up or re-auth with gh auth refresh -h github.com -s workflow.

Test plan

• Smoke test should pass: function now completes within 300 s
• Cold start: aggregate fetches in 30-65 s, full sitemap builds in 50-95 s, returns 750+ URLs
• ISR-warm start: aggregate from CDN in <1 s, full sitemap in ~20 s

🤖 Generated with Claude Code