designing-distributed-systems

Choose CP when correctness matters more than availability, such as banking balances, inventory counts, or seat booking. Choose AP when availability matters more, such as social media feeds, shopping carts, or product catalogs where stale reads and later conflict resolution are acceptable.

Leader-follower routes all writes through one leader and is simplest with strong consistency via synchronous replication. Multi-leader accepts writes in multiple datacenters but requires conflict resolution. Leaderless uses quorums (W + R > N) for maximum availability and partition tolerance.

Use sagas for distributed transactions across microservices because two-phase commit is slow, blocking, and reduces availability. Sagas coordinate via choreography (events) or orchestration (central coordinator) with compensating transactions for rollback.

A circuit breaker monitors failures and opens after a threshold, failing fast instead of waiting for timeouts. After a cooldown it enters half-open state to test recovery, closing again on success, which isolates failing services and protects the rest of the system.

Use causal consistency for chat applications so causally related messages (a reply and its original message) appear in the same order for all users. Concurrent messages may appear in any order, which is acceptable and avoids the latency cost of strong consistency.

Strong consistency requires synchronous coordination or quorum agreement before confirming writes. During a network partition, nodes that cannot reach a quorum must reject writes, making that portion of the system unavailable until the partition heals.