Consistency Models

A consistency model is a contract between a distributed system and its clients about what reads are allowed to see after writes happen.

The big three

• Linearizable — every operation appears to take effect at a single instant between its invocation and its response. Hardest to provide; easiest to reason about.
• Sequential — operations from all clients appear in some single total order, but the order doesn’t have to match real time.
• Eventual — reads can see stale data; given enough time without writes, all replicas converge.

CAP theorem

Eric Brewer’s CAP theorem says: in the face of a network partition, a system must choose between consistency and availability. There is no third option.

This is why most production systems are tunable: they offer linearizable reads with a latency cost, or fast eventually-consistent reads, and the client picks per request.

See Distributed Systems for the broader context.