Writes: the dirty secret of causal consistency

Causal consistency offers geo-distributed systems what ought to be a sweet option between the poor performance of strong consistency and the weak guarantees of eventual consistency. Yet, despite its appealing properties, causal consistency has seen limited adoption in industry, where systems have instead been clustering around the two extremes of eventual and strong consistency. We argue that this reluctance stems primarily from how causal consistency handles writes—both in how they are propagated, and in how conflicting writes are applied. We present our experience in designing and building two recent systems, Occult [25] and TARDiS [11], that try to address or mitigate these problems, and highlight some of the open challenges that remain in this space.