jsschools
Performance Tuning

Compress Responses, Keep Middleware Lean, and Benchmark with Autocannon

Performance Tuning

Apply koa-compress for gzip/br output, avoid blocking the event loop with sync work, and use autocannon to measure throughput before and after changes.

4 min read Level 2/5 #koa#production#performance
What you'll learn
  • Enable response compression with koa-compress
  • Identify and eliminate synchronous work that blocks the event loop
  • Run an autocannon benchmark to measure requests per second

Koa’s minimal core means performance is largely determined by what you add to it. A few targeted changes — compression, async I/O discipline, and lean middleware — can double throughput without architectural changes.

Response Compression

koa-compress adds Content-Encoding: gzip or br negotiation in a single middleware:

npm i koa-compress
import compress from "koa-compress";
import { constants } from "node:zlib";

app.use(
  compress({
    threshold: 1024, // skip compression for responses smaller than 1 KB
    gzip: { flush: constants.Z_SYNC_FLUSH },
    br: { params: { [constants.BROTLI_PARAM_QUALITY]: 4 } },
  })
);

Register compress early — before your router — so it wraps all downstream responses.

Avoid Synchronous Work

Node.js is single-threaded. Any synchronous CPU work blocks the event loop and stalls all concurrent requests.

AvoidUse instead
JSON.parse(bigString) in hot pathStream parsing, or move to worker
fs.readFileSyncfs.promises.readFile
crypto.pbkdf2Synccrypto.pbkdf2 (promisified)
bcrypt.hashSyncbcrypt.hash (async)

If you must do CPU-heavy work, offload it to a node:worker_threads worker so the event loop stays free.

Keep the Middleware Stack Lean

Every middleware function runs on every request. Audit your stack:

// Print the middleware count to catch accidental bloat
console.log("Middleware count:", app.middleware.length);
  • Mount route-specific middleware on the router, not the app.
  • Parse the body only on routes that need it — avoid a global body parser.
  • Move infrequent tasks (e.g., static file serving) behind a CDN instead.

Benchmarking with Autocannon

Install autocannon globally or as a dev dependency:

npm i -D autocannon

Run a 10-second benchmark with 10 concurrent connections:

npx autocannon -c 10 -d 10 http://localhost:3000/ping

Sample output:

┌─────────┬──────┬──────┬───────┬──────┬─────────┐
│ Stat    │ 2.5% │ 50%  │ 97.5% │ 99%  │ Avg     │
│ Latency │ 0 ms │ 0 ms │ 1 ms  │ 1 ms │ 0.21 ms │
├─────────┼──────┴──────┴───────┴──────┼─────────┤
│ Req/sec │ 28 000 – 42 000            │ 36 482  │
└─────────┴────────────────────────────┴─────────┘

Run the benchmark before and after each change to confirm the improvement is real. A 5–10% variance between runs is normal.

Up Next

Distribute load across all CPU cores using node:cluster and app.callback().

Clustering with node:cluster →