Eventual consistency is a model for replicated data in which the system guarantees that, if no new updates are made to an object, eventually all accesses will return the last updated value. Werner Vogels, then chief technology officer at Amazon, gave this widely cited definition in his 2008 article “Eventually Consistent,” drawing on the experience of building Amazon’s large-scale storage systems.
The key word is “eventually.” Immediately after a write, different replicas may still return different (older) values, and there is a window during which reads can be stale. Vogels frames eventual consistency as a specific form of weak consistency: the system does not promise that the next read sees the latest write, only that all replicas will converge once update activity stops and changes have had time to propagate.
This model is a direct response to the trade-off described by the CAP theorem. By relaxing the demand for immediate, system-wide consistency, a datastore can stay available and responsive even when nodes are partitioned or slow to communicate. That is why eventual consistency became the default behavior of many NoSQL and globally distributed systems that prioritize availability and low latency.
Vogels also notes that eventual consistency is not a single setting but a family. Variants such as read-your-writes, monotonic reads, monotonic writes, causal, and session consistency add targeted guarantees on top of the basic eventual promise, letting designers strengthen the model exactly where an application needs it without paying the full cost of strong consistency.