The Network & Latency
Rule: a healthy service plus a healthy service does not equal a healthy system. The network
Timeouts & Retries
Rule: retry without idempotency is not error handling — it is a bug. And a retry that fires
Circuit Breakers & Bulkheads
Rule: a dying dependency needs to be isolated, not attacked with more requests. Breakers stop
Consensus & Leader Election
Rule: agreement requires a majority, not a coordinator. Any scheme where a single node decides —
Designing for Failure
Rule: failure is the default state, not the exception. Large systems are not designed for the