Chaos Engineering
Deliberately injecting failure into a system to discover, before an incident does, which of your resilience assumptions are false.
Not "breaking things at random". The method is experimental: state a hypothesis about steady-state behaviour, define the smallest blast radius that can test it, inject one specific failure, and compare. Have an abort mechanism ready before you start.
The value is that resilience mechanisms are the least-exercised code in the system. Retry logic, failover, circuit breakers and DR runbooks all work in the design document; whether they work in production is unknown until something invokes them, and an incident is an expensive way to find out.
Start in a pre-production environment, start with a single instance termination, and only move to production once the basics stop finding problems — which usually takes longer than teams expect.