Why should you care?

CloudVerse checkout p95 must stay under 500ms during peak — profiling finds real limits.

See it live — copy this example

Use AWS/Azure/GCP free tier or local Docker/Kind. Sketches and YAML are meant to be typed and adapted.

# k6 load test snippet (CloudVerse checkout)
import http from 'k6/http';
import { check } from 'k6';
export const options = {
  stages: [
    { duration: '2m', target: 50 },
    { duration: '5m', target: 200 },
    { duration: '2m', target: 0 }
  ],
  thresholds: { http_req_duration: ['p(95)<500'] }
};
export default function () {
  const res = http.post('https://api.cloudverse.io/v1/checkout', JSON.stringify({ cartId: 'c-1' }), { headers: { 'Content-Type': 'application/json' } });
  check(res, { 'status 200': (r) => r.status === 200 });
}

What happened?

• Profile end-to-end: DB queries, cache hit rate, serialization, and network.
• Fix the largest slice of the latency budget.

Practice next

1. Baseline p95 with load test.
2. Profile slow trace span.
3. Apply one fix (index or cache).
4. Add Redis cache on catalog reads.
5. Enable HTTP/2 and connection pooling.

Remember

Measure before tuning. Fix biggest bottleneck. Set regression thresholds in CI.

CloudVerse checkout tune

p95 at 900ms under 200 RPS.

Outcome: DB index + cache cuts p95 to 420ms.