← knowledge.oriz.in

Cloudflare Worker quota mitigation playbook

decision cloudflareworkersquotascachingperformancedecisionsarchitecture

Cloudflare Worker quota mitigation playbook

Decision

Every Cloudflare Worker in the family follows a documented mitigation playbook to stay safely under the free tier. The user's concern was: "I'm using software workers for many more things, so I might outrun it." This decision codifies the techniques that keep the api.oriz.in umbrella Worker, the s.oriz.in short-link Worker, the Satori OG Worker route, and any future Worker safely below quota.

Free-tier limits to respect

The 8-step playbook

1. Cache the response with long Cache-Control

Set Cache-Control: public, max-age=31536000, immutable (or whatever freshness fits the resource) so the Cloudflare edge cache serves repeat requests without invoking the Worker. The Worker handles only the first request per cache key per edge node; every subsequent hit is served from cache and does not count.

Used by: Satori OG cards (parameter-hash key), short-link 302s (per-slug), any idempotent GET.

2. Use the Cloudflare Cache API inside the Worker

For dynamic responses where Cache-Control alone isn't enough:

const cache = caches.default;
const cacheKey = new Request(request.url, request);
let response = await cache.match(cacheKey);
if (!response) {
  response = await computeFreshResponse();
  ctx.waitUntil(cache.put(cacheKey, response.clone()));
}
return response;

Cache hits return early before any compute ? CPU stays near zero on hot keys. The waitUntil writes back asynchronously so the user gets the response immediately.

3. Move static logic to _headers / _redirects

Cloudflare Pages reads _headers and _redirects files at the edge without invoking a Worker. Anything that can be expressed as a static header (CSP, HSTS, CORS for known origins) or a static redirect (apex ? www, old paths ? new) belongs there, not in a Worker.

The security headers strategy already uses _headers. Apply the same posture to redirects.

4. Split Workers by domain — each gets its own 100K/day quota

The free-tier quota is per Worker, not per account. Splitting a single monolithic Worker into per-domain Workers multiplies the effective family-wide quota:

The umbrella Hono Worker decision keeps the API surface single — but routes that have their own domain (OG, short-link) are deployed as separate Workers and inherit independent quotas.

5. Use HTML Rewriter for cheap edge transforms

The HTMLRewriter API streams the upstream response through one or more transformers and counts as a single request — not "upstream + transform" billed separately. Use it for header injection, content rewrites, A/B variant swaps, and feature-flag gating instead of fetching, parsing, and re-emitting in Worker JavaScript (which would burn CPU).

6. CDN-cache OG images at the edge

The Satori OG endpoint returns a 1200×630 PNG keyed by parameter hash. Combined with #1 and #2, the Worker generates each unique title/theme combination once; every other request from any edge node is served from cache. With immutable cache headers, the realistic burn for a year of OG-card serving stays in the low thousands of generations even at million-pageview scale.

7. Use Workers KV for hot-path lookups

KV reads are faster than D1 / Neon and use less CPU. For per-request lookups (slug ? URL, license-key ? state, feature flag ? variant), KV is the right substrate. The 100k reads/day cap is per namespace; split namespaces per domain (same logic as #4).

8. Cron Triggers don't count against request quota

CF Cron Triggers run a Worker on a schedule and consume from a separate 1,000/day quota. Background work — RSS poll, idempotency sweep, cache rebuild, federation outbox flush — belongs on Cron Triggers, not on user-request handlers. The cron split decision already routes long / build-shaped jobs to GitHub Actions; CF Cron Triggers handle in-Worker periodic tasks.

Fail-safes (apply on top of the playbook)

Cross-refs