Causal Consistency

Alice posts "I got the job!" Bob replies "Congratulations!" Carol sees the reply but not the original post. That makes no sense, the reply references something Carol has not seen yet.

Causal consistency prevents this. If event B was caused by event A, anyone who sees B also sees A. The causal relationship is tracked (often with vector clocks) and the system ensures causally related events are observed in order.

Causal sits between eventual (too weak) and sequential (too strong). It is the sweet spot for many applications.

Causal consistency requires tracking which events depend on which. The common mechanisms:

Vector clocks, each node maintains a vector of logical timestamps. Events are causally ordered by comparing vectors.

Dependency tracking, each write carries metadata about which writes it depends on. A node only makes a write visible once all its dependencies are visible.

Explicit causal chains, application-level tracking. A reply carries the ID of the post it replies to. The system ensures the post is visible before the reply.