Reference · Glossary

Stress test

Pushing a system **beyond expected load** until it errors — finding breaking points and failure modes.

#When to use

Sizing concurrency limits, queue depths, and GPU memory for inference spikes.

Confusing stress tests with soak tests — stress finds ceilings; soak finds drift.

Ramp concurrent embedding jobs until OOM or p95 exceeds SLO — record max safe throughput.