Establishing a k6 Load Test Baseline

Without a baseline, “it feels slower” is not actionable.

Scenarios

// smoke.js — 10 VUs, 1 minute
export const options = { vus: 10, duration: '1m' };

// load.js — ramp to 500 VUs over 10m, hold 20m
export const options = {
  stages: [
    { duration: '10m', target: 500 },
    { duration: '20m', target: 500 },
    { duration: '5m', target: 0 },
  ],
};

Metrics We Track

• P50 / P95 / P99 latency
• Error rate (threshold: <0.1%)
• RPS at saturation point
• CPU/memory per pod at 500 VUs

CI Integration

Load tests run nightly against staging, not every PR (too slow). PRs run smoke only (~2 min). Release candidates run full load + 2-hour soak.

Baseline JSON checked into repo. Regression >10% on P99 blocks release until investigated.