How should availability and fault tolerance be applied in a production system?

Define SLOs, remove critical single points, spread across failure domains, automate failover, and test failure behavior.