load-test

Define scenarios using the constant-arrival-rate or ramping-arrival-rate executor with a target RPS, then model realistic user journeys with groups, checks, and think time. Add thresholds for P95 latency and error rate so the test fails automatically when SLOs are violated.

k6 uses JavaScript scripts with code-first configuration, built-in thresholds, and easy CI integration, making it well suited for developer workflows. JMeter uses XML test plans and a GUI, which fits Java enterprise environments with existing JMeter infrastructure.

Use arrival rate rather than VU count. Constant arrival rate generates a fixed number of requests per second regardless of response time, which models real traffic more accurately than a fixed pool of threads that slows down when the system degrades.

Never run load tests against production without explicit traffic isolation. Use a production-like staging environment, start at 10% of target load to baseline, and ramp up gradually while monitoring database and resource utilization.

Common causes include unrealistic traffic models that hammer one endpoint, missing think time, no spike or soak scenarios, and unmonitored database bottlenecks. Model the real request mix, add 1-5 second pauses, and run soak tests to catch memory leaks.