Load Testing
Driving a system with realistic traffic at a target volume to verify it meets its performance targets before real users do.
Distinguish the varieties, because they answer different questions. Load testing checks expected volume. Stress testing increases load until something breaks, which is how you learn what breaks first and whether it fails gracefully. Soak testing runs for hours or days to surface leaks and slow degradation. Spike testing checks the response to a sudden jump, which is what autoscaling and cold starts fail.
Most load tests mislead for the same three reasons: the data volume is unrealistically small so everything is in cache, the request mix is uniform so there is no hot key, and the environment is smaller than production in a way that changes the bottleneck rather than scaling it.
The output worth having is not a pass mark. It is a saturation curve — throughput and latency against concurrency — and a named first bottleneck.