A chaos testing tool for your existing tests.
Chaos Lab in Nexus StudioQA injects a fault into the browser while one of your tests runs: network latency, an offline window, a slower CPU or a failing dependency. The run ends CHAOS_RESILIENT if the test held, CHAOS_BREACHED if it did not, and CHAOS_NOT_INJECTED when the fault never took effect, which says nothing about resilience.
By the Nexus StudioQA team, Verdict System LLC.
Last updated: .
The fault modes
| Mode | What it does | Default |
|---|---|---|
| network | Holds every request for a set time; about 20% get a few hundred more milliseconds | 400 ms |
| offline | Cuts the network once, at a random moment in the first 30 seconds | 5000 ms |
| cpu | Slows the CPU by a factor of 4; Chromium only | 4x, fixed |
| all | Network, CPU and offline together | Defaults above |
| dependency | Answers one URL you name with 429, 500, 503, slow, hang or a cut body | Set by you |
Packet loss and memory pressure can ride on top of any mode. A dependency fault also watches for a retry storm: more than 10 requests to the faulted URL inside 10 seconds. You run a fault with nexus patrol TC01 --chaos=network, or from the Chaos Lab screen.
How a chaos run is judged
| Verdict | Meaning |
|---|---|
| CHAOS_RESILIENT | The test passed with the fault in effect |
| CHAOS_BREACHED | The test failed under the fault |
| CHAOS_NOT_INJECTED | The test passed, but no fault took effect, for example cpu on Firefox. It is not a result and is left out of every rate |
The verdicts page has the full rules. Chaos runs are filed in their own evidence folder, so an injected failure never counts against the stability of the real suite.
Steady state, not only assertions
A test can pass and still be much slower under the fault. Chaos Lab compares each run with the median duration of the same test and browser without a fault. A run slower than 2 times that baseline is marked as deviating (the tolerance is NEXUS_CHAOS_TOLERANCE). It also records the Web Vitals measured under the fault. Over time, each test gets a resilience label: RESILIENT at 90% or more, TOLERANT at 70%, FRAGILE at 50%, VULNERABLE below that.
Infrastructure faults
nexus chaos-infra <target> --action=pause|stop|kill|restart --for=<seconds> takes a registered Docker container, or the app Nexus launched, down on purpose and measures whether it comes back: RESILIENT, DEGRADED or NOT_RECOVERED. It needs authorization before it acts, and --restore undoes a fault left in.
What it does not cover
API tests run without chaos. Selenium on Firefox is skipped under chaos, because it would pass with no fault. --chaos and --visual cannot be combined.
Common questions
Do I need to rewrite my tests for chaos?
No. The fault is applied around the tests you already have, in Playwright, or in Python and Selenium on Chromium. Run each test once without a fault first, so the steady-state check has a baseline.
What does CHAOS_NOT_INJECTED mean?
The test passed, but the fault never took effect. For example, CPU throttling works only in Chromium. It is not counted as resilient or breached, so it cannot inflate or lower a rate.
Can chaos break my production system?
Browser faults act on the page under test, not on your servers. Infrastructure faults act on containers you register, ask for authorization first, and can be undone with a restore. Do not point either at a system you do not own.
Does a chaos failure count as a bug?
It is a finding about resilience, filed apart from the normal suite. It does not change the stability numbers of the real tests.
Find out what happens when things go wrong.
Everything runs on your own machines. No card needed.