Generate K6 Load Test Scripts and Analysis

This skill writes k6 load test scripts - JavaScript API, test design patterns, metrics analysis, and performance optimization. Core principles: Virtual Users (VUs) each run the script independently in parallel, tests are either iteration-based or duration-based, stages gradually ramp load up and down to simulate realistic traffic, thresholds define automated pass/fail criteria, and analysis focuses on response time, throughput, and error-rate metrics.

Use it when a test needs a realistic staged ramp-up/ramp-down pattern with automated thresholds, not a single fixed-VU run.

export const options = {
  stages: [
    { duration: '2m', target: 10 },
    { duration: '5m', target: 10 },
    { duration: '2m', target: 20 },
    { duration: '5m', target: 20 },
    { duration: '2m', target: 0 },
  ],
  thresholds: {
    http_req_duration: ['p(95)<500'],
    http_req_failed: ['rate<0.1'],
  },
};

A standard script defines a setup() that runs once to prepare test data or authenticate, a default exported function that runs per-VU per-iteration with check() assertions and sleep() think-time, and a teardown() for cleanup. Advanced patterns include data-driven testing with SharedArray loading a CSV via papaparse to avoid duplicating large datasets across VUs, and session-based testing using group() blocks to structure an Authentication phase (capturing a token) followed by a User Operations phase reusing that token across profile-read and profile-update requests. Two named load patterns are given directly: spike testing (a 10-second ramp to 100 VUs, hold, quick ramp-down) and stress testing (staged increases from 100 to 300 VUs to find the point beyond normal capacity). Environment configuration covers CLI flags (--vus, --duration, --env, --out influxdb=..., --out json=...) and reading __ENV.BASE_URL/__ENV.API_KEY inside the script with defaults.

Key metrics to monitor are named explicitly - http_req_duration (p50/p95/p99), http_req_failed, http_reqs (throughput), vus, and iterations - alongside custom Counter and Gauge metrics tagged by endpoint and version for finer-grained analysis. handleSummary() can emit results to stdout via textSummary, to summary.json, and to an HTML report simultaneously.

Performance engineers writing k6 load tests who need concrete staged-ramp, spike, and stress patterns plus custom metrics rather than a single default script. Eight optimization tips close it out: use SharedArray for large datasets to avoid per-VU memory duplication, implement realistic think time with sleep(), batch HTTP requests where possible, set appropriate timeouts, reuse connections instead of recreating them, monitor the load generator's own resource usage, ramp up gradually rather than overwhelming the system immediately, and use tags and groups for clearer result analysis and debugging.