Eventual Consistency
A primer on distributed state that converges over time.
What it is
Eventual consistency is a guarantee that if no new updates are made to a given data item, all reads will eventually return the last updated value. It trades immediate correctness for availability and partition tolerance under the CAP theorem.
Why it matters
In globally distributed systems, synchronous replication adds latency that degrades user experience. Eventual consistency allows writes to succeed locally and propagate asynchronously, keeping response times low while the system self-heals.
Conflict resolution
When concurrent writes collide, the system must reconcile differences. Common strategies include last-writer-wins (LWW) with wall-clock or logical timestamps, CRDTs that merge deterministically, and application-level merge functions.
Real-world examples
- DNS — zone changes propagate over TTL windows.
- Amazon DynamoDB — configurable consistency levels per request.
- Meridian session tokens — signed, cached locally, verified against an eventually-consistent revocation set.
When to avoid it
Strong consistency is required for financial ledgers, inventory decrements that must never oversell, and any domain where a stale read causes irreversible harm. In those cases, use quorum writes or a linearizable datastore.