Even with the best-designed DR strategy, "does it actually work?" is separate. The 5-minute RTO on paper can become 30 minutes when real disaster strikes, and systems touted as "Multi-AZ, so we're fine" crumble when an AZ actually dies. The uncomfortable truth of software reliability: "untested recovery doesn't work." Enter Chaos Engineering—intentionally injecting failures into healthy systems to surface weaknesses before they strike for real.
AWS commercialized this via two services