What Eventual Consistency Means
In a distributed store, a write may be accepted on one replica and copied to others a moment later. Eventual consistency means that if writes stop, all replicas will converge to the same value, but for a brief window a read can return stale data from a replica that has not caught up.
Why Systems Choose It
- It keeps the system available even when replicas are slow or partitioned.
- Reads can be served from the nearest replica for low latency.
- It avoids the coordination cost of making every read see the very latest write.
The Tradeoff in Practice
A user who updates a profile then reloads might briefly see the old value if the read hits a lagging replica. This is the same read your own writes problem seen in replication, now built into the consistency model.
Tools to Manage It
- Choose strong reads for the few operations that cannot tolerate staleness.
- Route a user to a replica that has seen their write when freshness matters.
- Design features so brief staleness is harmless, like a like count that settles.
Key idea
Eventual consistency trades momentary staleness for availability and low latency, so reads may return old data until replicas converge.