Cold starts & scaling

Nimbus functions run in isolates, not containers, which is what keeps cold starts under 5ms — there’s no OS or runtime to boot, only your code.

How scaling works

Each region scales the number of running isolates for your function independently, based on concurrent request volume in that region. There’s no configuration for this: it happens automatically, and you’re billed per request rather than per running instance.

Tip

If a function does meaningful work on every cold start (for example, opening a database connection), keep that setup code outside the request handler so it only runs once per isolate, not once per request.

functions/db.ts

import { connect } from '@nimbus/db';

// Runs once per isolate, not once per request.
const db = await connect(process.env.DATABASE_URL);

export default async function handler(request: Request) {
  const rows = await db.query('select 1');
  return Response.json(rows);
}

Concurrency limits

Free-tier projects are capped at 100 concurrent requests per region. Paid plans raise this limit; see nimbus billing plans for current numbers.