recall

← recall

CAP theorem conceptpaper

During a network partition between nodes, a distributed system must choose between consistency (every read sees the latest write) and availability (every request gets a response).

During a network partition between nodes, a distributed system must choose between consistency (every read sees the latest write) and availability (every request gets a response). Outside of partitions, the tradeoff doesn't apply — that's the famous misstatement. CAP is binary and pessimistic; the practical follow-on PACELC adds latency-vs-consistency for the no-partition case. CP systems (Spanner, ZooKeeper) refuse requests during partitions. AP systems (Dynamo, Cassandra) keep serving but may diverge. The choice is per-operation, not per-system in modern designs.

Often misstated. The real claim: when a network partition happens, you must choose between consistency and availability. Most systems are CP or AP under partition.

see also

aliases: CAP

topics: distributed-systems

references: