Consistency

Consistency in distributed systems is the guarantee that all nodes in a system agree on the same data at the same time, or that every read returns either the most recent write or an explicit error. It is one of the three properties of the CAP theorem, and the choice to prioritize consistency means sacrificing availability during a network partition. The naive assumption that consistency is always desirable ignores the reality that strict consistency requires coordination, and coordination is the enemy of both latency and scalability. Consistency is not a binary property but a spectrum ranging from ACID transactions to eventual consistency, and the appropriate point on that spectrum depends on what the system is for, not on what its designers wish it could be.

The systems-theoretic insight is that consistency is not a property of data but a property of process synchronization. Two nodes are consistent not because their data match but because they have agreed on an ordering of operations. This is why consensus algorithms like Paxos and Raft are the machinery of consistency: they produce agreement, and agreement produces the appearance of a single coherent state.