CPU Time and Database

Where to tune the Worker: the CPU limit, how far the database is, and Hyperdrive's query cache.

Three settings decide most of what a request costs. Change one only for a reason you measured in Workers Logs or Cloudflare's metrics.

CPU Time per Request

worker.limits.cpu_ms in apps/web/void.config.ts sets how much CPU time one request may use, from 1 to 300000 ms.

export default defineConfig({
worker: { limits: { cpu_ms: 30000 } },
});
  • Workers Paid allows up to 300000 ms; Workers Free only 10 ms, too little for password hashing. The kit needs Paid (Deployment).
  • It is CPU time, not waiting: waiting for the database or a fetch is free.
  • Raise it only when a request is cut off for CPU. If one route needs minutes of CPU, it belongs in a queue job (Background Jobs), not a request.

Where the Database Is

Every query is a round trip from the Worker to your database, and Hyperdrive keeps connections to it pooled and warm. A far database costs more than anything in your code.

  • Create the database in the region closest to most of your visitors.
  • For pages that run several queries, try Smart Placement (Cloudflare dashboard → your Worker → Settings → Placement): it runs the Worker near the database. Compare before and after; it can make a page with one query slower (Placement).
  • Count queries per page in a loader before anything else. A request has one database connection, so queries run one after another even inside Promise.all: fewer queries (a join, one combined query) is what saves round trips.

Hyperdrive's Query Cache

Hyperdrive caches SELECT results for up to 60 seconds by default, and void deploy leaves it on. For this app it is off by design: a cached read keeps a revoked session, or a member removed from an organization, alive for up to a minute. Turn it off once after the first deploy (Production Deploys) and check it in the dashboard (Hyperdrive → your config → Caching). The health check is meaningful only with the cache off.

Connections

The deployed Worker connects through the HYPERDRIVE binding. DATABASE_URL is for deploys and local work and is never a Worker secret (Production Deploys).