Simple systems tend to last.

Designing for graceful recovery

Reliable services are built around clear failure boundaries. A restart should be predictable, configuration should be inspectable, and recovery should require as few moving parts as possible.