Load Testing vs Stress Testing vs Soak vs Spike
Load testing vs stress testing, soak and spike tests: what each asks, the load shape behind it, what it finds, and the order to run them on an API.

A load test runs the traffic you expect. A stress test pushes past it until something gives. A soak test holds normal traffic for hours. A spike test jumps to a burst with almost no ramp. Start with a load test, then pick the next one by the risk your API carries. The rest of this page gives each test its question, its load shape and what it finds, then says when to run it.
The four tests in one table
The load testing guide covers how to build and read a single test. This page compares the four shapes. The names follow the k6 test types guide, and they differ between teams: a stress test for one application is a plain load test for another.
| Test | What it asks | Load shape | What it finds |
|---|---|---|---|
| Load | Does the API hold at the traffic you expect? | Ramp to the expected level, hold, ramp down | The baseline: p95, error rate and requests per second |
| Stress | What happens past that traffic? | The same pattern as a load test with a higher target | Where latency bends and what breaks first |
| Soak | Does it stay healthy for hours? | Normal load, held for hours | Leaks, filling disks and queues, slow drift |
| Spike | Does it survive a sudden jump, and how fast does it recover? | A very short or non-existent ramp-up to an extreme level | Survival and recovery time |
Load testing: the baseline
A load test applies the traffic you expect and records how the API behaves. It is the reference for every other test, because a stress, soak or spike result only means something next to the normal-load numbers. Before it, run a smoke test: a run with a few users that proves the flow and the target work before you spend a real run. After that, the load test gives you a trustworthy baseline for p95 latency, error rate and requests per second.
Stress testing in software testing
Stress testing in software testing means loading a system beyond its normal level to see how it behaves. The shape is a load test with a higher target, and the k6 stress testing guide calls it testing with loads “heavier than usual”.
How much higher is your call. Some testers aim for 50 or 100 percent more than the normal load, but there is no fixed percentage. Pick the pressure your API could actually meet.
Read the ramp. Past normal traffic, latency usually climbs first, and at some level errors start or the service falls over. The useful output is the point where latency starts to bend and the first thing that breaks. Run stress tests only after load tests have passed.
Soak testing: the long run
A soak test keeps normal load on the API for a long time. It looks for problems that only show after hours of use: responses that slowly get slower, memory that is never released, disks and queues that fill up.
The k6 soak testing guide suggests holding the load for hours or even days, with typical values of 3, 4, 8, 12, 24, and 48 to 72 hours. Watch the server while it runs. The API’s latency chart shows the drift but not which resource is filling, so track memory, disk and connection counts as well.
Stresseur limits apply here. A single load test runs for up to 10 minutes on Free and up to 1 hour on Team. Scheduled soak tests are not available. The shortest typical soak duration the k6 guide lists is 3 hours, which is beyond both limits.
Spike testing: the sudden surge
A spike test checks whether the API survives a burst. The load jumps to an extreme level with almost no ramp.
It matters when real traffic arrives in bursts. The k6 guide lists ticket sales, product launches, broadcast ads, process deadlines and seasonal sales as typical causes.
Read recovery as well as survival. After the burst ends, note how long latency and errors take to return to their normal level. An API that survives the spike but is slow to return to normal still has a problem to fix.
Which test to run first
Start with a smoke test, then a load test, then pick by risk.
- Traffic grows or arrives in rushes: stress test.
- Failures appear after hours of uptime: soak test. Do not run it before the smoke and load tests.
- Launches, broadcast ads or deadlines: spike test.
Give every run a pass or fail rule. Stresseur profiles support thresholds and abort rules, so each run can carry its own pass or fail rule. Without a rule, the result is a chart and an argument. For the next question, how many virtual users to start with, see choosing virtual users for an API. To see how these tests sit inside the wider discipline, read load testing and the broader discipline.
Running them on a recorded flow
In Stresseur any recorded flow can also run as a load test. Without a profile file you get Smoke, Load, Stress, Spike, Soak and Breakpoint, and the executors are constant-vus, ramping-vus, constant-arrival-rate and ramping-arrival-rate. Load generators start in the regions you pick (Frankfurt, Virginia, San Jose, Singapore), and results stream live: virtual users, req/s, p95, errors and per-step latency.
The load testing page shows a flow running under load, and the load test docs hold the profile reference. Free allows load tests up to 50 VUs for 10 minutes in 1 region, and Team up to 1,000 VUs for 1 hour in all regions.
k6 is the better fit for a soak run of that length: its guide’s soak example holds 100 users for 8 hours. A recorded flow is the better fit when the load test should reuse a flow you already have, as in multi-step flow load tests.
Conclusion
Load testing sets the baseline. Stress testing finds the ceiling, soak testing finds what fails with time, and spike testing finds what fails on sudden demand. Run smoke and load first, then choose the next test by the risk your API carries, and give each run a pass or fail rule.
Frequently asked questions
What is the difference between load testing and stress testing?
A load test runs the traffic you expect and records the baseline. A stress test uses the same shape with a higher target, to find what happens once traffic goes past that level and what gives way first.
How long should a soak test run?
Long enough for slow problems to show, which means hours. The k6 guide suggests 3, 4, 8, 12, 24, or 48 to 72 hours. On Stresseur a single load test is capped by plan, at 10 minutes on Free and 1 hour on Team.
Is spike testing the same as stress testing?
No. A stress test climbs past normal traffic in a ramp and shows where the API bends. A spike test skips the ramp and jumps to a burst almost at once, which shows whether the API survives it and how fast it recovers.
Can Stresseur run a soak test?
Stresseur has a Soak profile among its defaults. Scheduled soak tests are not available, and a single run is capped at 10 minutes on Free and 1 hour on Team.