fix(ci): cap V8 heap below Cloudflare's 8 GB build container

[dcrawbuck]

Problem

Cloudflare Workers Builds (the git-connected CI that builds & deploys the docs Worker on push to main) has been intermittently failing on main. Builds end with:

[prerender] Prerendering pages...
[prerender] Crawling: /
An internal error occurred. Please retry your build. If this problem persists, contact Cloudflare support.

(outcome: terminated). The client and SSR bundles build fine — the build dies the instant the prerender crawl starts.

Root cause

The build runs NODE_OPTIONS=--max-old-space-size=8192 — an 8 GB V8 heap — but Workers Builds containers have exactly 8 GB of RAM (20-min timeout, so this is not a timeout). Telling V8 it may grow to 8 GB on an 8 GB box defers the major GC until RSS already exceeds the container, so the kernel OOM-kills the build during the prerender crawl (its peak-memory moment). The abrupt stop with no stack trace — just Cloudflare's "internal error" — is the OOM-kill signature.

GitHub PR CI never caught this because its runners have 16 GB, so the same 8 GB heap fits.

Fix

• Lower the V8 heap ceiling to 5 GB (--max-old-space-size=5120) in the build and build:cf:staging scripts — above the ~4 GB the SSR bundle needs (4096 flaky-crashes with V8 "heap out of memory"), and below the 8 GB container so the major GC fires before RSS hits the ceiling.
• Drop prerender concurrency 2 → 1 to trim the crawl's peak memory.

Prerendering stays fully enabled; site output is byte-for-byte identical.

Verification

• Local build with the shipped config: 556 pages prerendered, bun test 65 pass / 0 fail, ~2.2 min.
• Confirmed the 8 GB container limit (CF docs), real-node heap-cap enforcement, and the ~4 GB SSR floor.
• Final 8 GB-container proof: this PR's Cloudflare preview build.

Same fix also covers staging (shares the build script). Do not merge until the preview build is green.

---

Note

Low Risk
Build-time memory tuning only; no runtime app, auth, or data-path changes, and prerender output is intended to stay identical.

Overview
Lowers the Node/V8 old-space heap limit for production docs builds from 8 GB to 5 GB (--max-old-space-size=5120) on the main build script so prerender can run inside Cloudflare’s 8 GB RAM build containers without the kernel OOM-killing the process when the heap matches container size.

The build:cf:staging script no longer sets its own NODE_OPTIONS=8192; it only sets CLOUDFLARE_ENV=staging and delegates to build, so staging uses the same 5 GB cap.

^{Reviewed by Cursor Bugbot for commit 5272278. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	superwall-docs	5272278	Commit Preview URL Branch Preview URL	Jun 25 2026, 04:57 PM