workers-best-practices

Check code against known anti-patterns: buffered unbounded response bodies, hardcoded secrets, Math.random() for tokens, floating promises, and module-level mutable state. Verify wrangler config uses current compatibility dates, nodejs_compat, and generated Env types from wrangler types.

Set observability.enabled and observability.traces.enabled to true in wrangler.jsonc; the top-level setting alone does not enable traces. Use head_sampling_rate to control volume and emit structured JSON logs via console.log(JSON.stringify({...})).

Workers have a 128 MB memory limit, and buffering entire bodies with await response.text() or response.arrayBuffer() exhausts it on large payloads. Stream data through using TransformStream or pass response.body directly to the new Response.

Use bindings for Cloudflare services like KV, R2, D1, and Queues instead of REST API calls. Bindings are direct in-process references with no network hop or authentication overhead, while REST calls add latency and credential management complexity.

Module export handlers like fetch and scheduled access bindings via the env parameter, while platform base classes like WorkerEntrypoint, DurableObject, and Workflow access them via this.env.X. Using the wrong pattern is the most common binding error.

Use ctx.waitUntil() for background work like analytics, cache writes, or webhooks that should complete after the response is sent, with a 30-second post-response limit. Never destructure ctx, as it loses the this binding and throws Illegal invocation.