Replication Lag
Summary: The delay between a write being accepted by the leader and its arrival at a follower.
Sources: chapter5
Last updated: 2026-04-18
Replication lag is a fundamental characteristic of eventual-consistency. In an ideal world, the lag is only a fraction of a second, but in practice, it can increase due to network congestion, load on the followers, or large write volumes.
Concurrency Anomalies
High replication lag can lead to several user-facing issues:
- Reading your own writes: A user makes an update and immediately reloads the page, but the load balancer sends the request to a lagging follower, making it appear as if the update was lost (source: chapter5, p. 162).
- Moving backward in time: A user refreshes a page multiple times and sees state that flip-flops between new and old as different followers are queried.
- Violating causality: A user sees an effect (e.g., a reply to a comment) before they see the cause (the original comment).
Consistency Guarantees
To mitigate replication lag, systems can provide: