Prerendered docs page routes 404 under local wrangler dev / opennextjs-cloudflare preview, so no page can be smoke-tested in the Workers runtime before merge #265

Activity

1. os-project-manager

   added the

   Task

   issue type on Sep 4, 2026
2. os-project-manager

   added

   pm:queue

   on Sep 4, 2026
3. os-bill commented on Sep 8, 2026

   

   os-bill

   on Sep 8, 2026 – with Claude

   Collaborator

   More actions

   Closing completed — mechanism established, fixed, and the fix is on main

   repo:objectos seat, session_01ChPQM8jamxLUfUAxwFpJ8S (new tenure, maintainer-ordered forced takeover — audit on objectstack#9831). This card had zero comments: the fix landed but nothing was ever written here, so a reader of the card alone would still think the gap was open.

   This card's acceptance bar was explicit and modest: "First task is to establish the mechanism, and reporting a measured cause with no fix is a complete delivery." That bar is cleared, and then some — the cause was established and fixed.

   The mechanism — and it turned out to be the same defect that took the site down

   Not one of the four directions this card listed as guesses. defineCloudflareConfig() was being called with no arguments, which resolves incrementalCache to "dummy" — a cache whose get() throws on every call by design. Every prerendered-page lookup therefore missed. Combined with dynamicParams = false on the docs route, the miss fell through to NoFallbackError → _not-found → 404 on every page.

   That is why /docs, /docs/build/automation/approvals and /zh-Hans/docs/... all returned the same 11455-byte 404 body while /api/search, /llms.txt, /sitemap.xml and /og/* worked: the working routes are the ones that never consult the incremental cache.

   It was never a preview-tool quirk. It was a real, shipped defect that local preview was faithfully reproducing — and the same one that produced the 2026-09-04 production outage. This card's discipline is what makes that legible: it recorded the observation, refused to assert a cause it had not measured, and listed its guesses as guesses. All four guesses were wrong, and naming them as unmeasured is what kept them from becoming the finding.

   What I verified myself, and what I am taking on someone else's measurement

   ⚠️ Scoping this honestly rather than presenting it as one verification.

   Verified by me, directly on origin/main @ 0e26657:

   apps/docs/open-next.config.ts:70  export default defineCloudflareConfig({
   apps/docs/open-next.config.ts:71    incrementalCache: staticAssetsIncrementalCache,
   apps/docs/open-next.config.ts:72  });
   

   The cache is configured — the precise inverse of the root cause. The file also carries the failure mode written out at :13-19, so the next reader meets the explanation at the point of use rather than in an issue thread. Control-probed the pathspec (it returns other cloudflare hits in apps/docs), so this is a real reading.

   Inherited, not re-run by me: the behavioural half — node .github/scripts/smoke-docs.mjs --base http://localhost:8792 returning ✓ smoke: 4 page(s) rendered, negative control demonstrated red on the fix branch, and the same command producing 21 findings against main at d6f5dda. That is recorded in #274 and in objectstack#9831 §5. I did not re-execute opennextjs-cloudflare build && preview; I am closing on a structural check plus that record, and saying so.

   What this unblocks

   Local preview now serves real prerendered pages under real workerd with no Cloudflare credentials, so the pre-merge rendering check this card's absence made impossible is now buildable. That work is #274, pm:queue, and is the remaining half: rendering is still verified only after deploy, so the first observer of a broken render is a live visitor.

   pm:queue cleared in the same stroke as the close — a closed card keeps no pm:* state.

   ---

   Generated by Claude Code
4. os-bill

   removed

   pm:queue

   on Sep 8, 2026
5. os-bill

   closed this as completedon Sep 8, 2026