Load Test Configurator

k6 is a modern load testing tool written in Go with a JavaScript scripting API. It's performant, runs locally or in the cloud, and has excellent Grafana integration. Artillery is a Node.js-based load tester with a YAML-first config approach and a plugin ecosystem. k6 is better for high-throughput testing and CI integration; Artillery is easier for non-developers to configure and has better built-in reporting for multi-step scenarios.

VUs (virtual users) simulate concurrent users. Each VU runs your test script sequentially — making requests, waiting for responses, then making more. RPS (requests per second) is the throughput. The relationship is: RPS = VUs / response_time_seconds. With 100 VUs and 200ms average response time, you get ~500 RPS. Most load tests are configured by VU count; RPS is the result you measure.

A ramp-up period gradually increases load from 0 to the target VU count over a set time. This mimics real traffic patterns and gives your application time to warm up (JVM JIT compilation, connection pool initialization, cache warming). Without ramp-up, a sudden spike of 1000 VUs can cause artificial errors from cold-start conditions that wouldn't occur in production.

Common industry thresholds: p95 response time < 500ms for APIs, p99 < 1000ms; error rate < 1% under normal load, < 5% under stress load. For web pages: Time to First Byte (TTFB) < 200ms. The exact values depend on your SLA and user expectations. Define thresholds before testing — deciding after seeing results introduces confirmation bias.

Stress test: gradually increase load until the system fails — finds the breaking point. Spike test: instantly jump to 10x normal load — tests elasticity and auto-scaling behavior. Soak test: run at normal load for 24+ hours — finds memory leaks, connection pool exhaustion, and drift in performance over time. Load test: run at expected production load — validates performance SLAs. Use all four for comprehensive coverage.