Performance Testing

Performance testing validates system behavior under various load conditions, from normal operational levels to breaking points. Unlike functional testing, performance testing finds non-functional defects — too slow, crashes under load, degrades over time. The five primary test types (load, stress, soak, spike, and volume) each probe different failure modes. Performance budgets — p99 latency <200ms, error rate <0.1% — translate non-functional requirements into enforceable CI gates using tools like k6 with thresholds.

Key Points

• Load testing: simulate expected concurrent users and request rates to verify system meets SLOs under normal operating conditions.
• Stress testing: increase load beyond expected maximum to find the breaking point — identifies at what concurrency errors spike and which component fails first.
• Soak testing (endurance): run at moderate load for 24–72 hours — reveals memory leaks, connection pool exhaustion, and slow disk fill that only appear over time.
• Spike testing: sudden 10x traffic burst for 30–60 seconds — validates auto-scaling speed, queue back-pressure, and graceful degradation under sudden demand.
• Volume testing: process extremely large data sets — validates DB query plans don't degrade, exports don't OOM, and batch jobs complete within SLA windows.
• Baseline benchmarks: establish p50/p95/p99 latency and throughput at N concurrent users — run after every major release to detect performance regressions.
• k6 thresholds: fail test if `http_req_duration{p(99)} > 200ms` or `rate(http_req_failed) > 0.01` — enforce performance SLOs as CI gates.
• Distributed load generation: single machine tops out at ~10,000 concurrent connections — k6 Cloud, Locust distributed mode, or Artillery Cloud for higher concurrency.