Cloudflare Workers

Workers run JS/TS at 300+ edge locations with sub-millisecond cold starts. Trade-off: a constrained runtime (no Node APIs, 128MB RAM, 30s CPU on paid plan, 50ms on free). The right shape: thin compute that fans out to bindings (KV, D1, R2, DO, Queues).

When to use

• Stateless API at the edge — auth proxy, GraphQL gateway, image transforms.
• Static + dynamic hybrid — Pages for static, Worker for dynamic routes.
• Real-time at scale — Durable Objects for per-room state.
• Cost-driven migration from Lambda + DynamoDB.

When NOT to use

• Heavy CPU work (ML inference, video processing) — buy compute elsewhere.
• Long-running connections (>30s WebSocket from a regular Worker) — use DO with hibernating WebSockets instead.
• Apps that need full Node ecosystem (e.g. Puppeteer) — different runtime.

The bindings model (key insight)