Design for failure -- everything will break eventually

Systems designed assuming everything works are fragile. A single failure cascades into total outage. Developers do not consider failure modes during design.

For every component, ask: what happens when this fails? Design answers: (1) Timeouts on all external calls. (2) Retries with backoff and jitter. (3) Circuit breakers for repeated failures. (4) Fallback values for non-critical data. (5) Bulkheads to isolate failure domains. (6) Health checks and automated recovery. (7) Graceful degradation: show cached data, disable features, queue for later. (8) Chaos testing to verify failure handling works.

In distributed systems, failure is not exceptional -- it is normal. Networks partition, disks fill, processes crash, DNS fails. Systems that handle failure gracefully are the ones that stay up.