Engineering Craft › Testing

Stress Testing

Pushing beyond capacity to find the breaking point.

Also known as: stress testing, stress test, load to failure

A stress test pushes a system past its expected capacity, to find where it breaks and how. A normal load test asks “does it meet the target under expected load?”; a stress test asks “what happens at 2×, 5×, 10× — and does it fail safely?” Finding the breaking point is the goal, not avoiding it.

What you learn:

• Where the limit is — the throughput or concurrency at which latency spikes or errors begin.
• How it fails — does it degrade gracefully, rejecting excess work (backpressure, load shedding), or collapse entirely? A good system sheds load; a bad one falls over and stays down.
• Whether it recovers — when the load subsides, does it return to normal, or is it left wedged (threads stuck, a queue still full, a circuit breaker not resetting)?

ramp: 1× → 2× → 4× → 8× expected load
watch: latency, error rate, queue depth, CPU/memory, recovery after

The classic mistakes:

• Stopping at “it broke”. The interesting information is how it broke and whether it recovered. A system that rejects excess load cleanly is very different from one that takes itself down.
• No recovery check. Let the load fall away and watch. Many systems fail once and never come back without a restart — a worse outcome than the original overload.
• Confusing it with a load test. Testing at expected load is a load test; testing beyond is a stress test. Both matter, for different questions.
• Running it against production. Deliberately overloading a live system is almost always wrong. Use an isolated environment sized like production.
• Only testing the happy path. Stress the parts that fail first: the database connection pool, a shared queue, the one synchronous dependency.

Stress testing is how you learn the shape of failure before you meet it at 3am. Knowing that your system sheds load at 3× and recovers in 30 seconds is far better than discovering it can’t recover at all. Pair it with capacity planning (where you want the limit to be) and soak testing (whether it stays healthy over time).