Flash card
AWS Fault Injection Service: managed fault injection against AWS resources, run as a controlled experiment.
- Runs experiment templates that inject real faults: instance termination, API throttling, CPU and memory pressure, network latency, Availability Zone impairment.
- Every template carries stop conditions wired to CloudWatch alarms, so an experiment that breaches your own guardrail halts itself.
- Scoped by resource tags and filters, which is how a blast radius is kept to the workload under test.
- Actions run through IAM, so what an experiment can touch is bounded by a role rather than by care.
- Its output is evidence about recovery behaviour, not a recommendation: it tells you what happened, and you decide what to change.
Pick it when
Pick it when the requirement is to prove a recovery path works, rather than to find inefficiency. It is the service that answers “does the failover actually happen, and how long does it take”.
It's the wrong answer when
It is the wrong answer for finding problems you have not thought of, and for anything in the cost or right-sizing space. It tests hypotheses you supply. A scenario asking what to improve wants Compute Optimizer, Trusted Advisor, or the Well-Architected Tool; this one asks whether what you built survives.