Problems with Replication Lag

In a "read-scaling" architecture (many followers), you typically use asynchronous replication. This leads to Eventual Consistency: if you stop writing, all followers will eventually catch up. But while they are lagging, you can see strange things.

1. Reading Your Own Writes

If you update your profile and immediately reload the page, you might hit a follower that hasn't received the update yet. It looks like your update was lost! Solution: Read-After-Write Consistency.

• Read from the leader for things the user just modified (e.g., their own profile).
• Track the timestamp of the last update and read from a replica that is at least that up-to-date.

2. Monotonic Reads

If a user reads from a fresh replica, then refreshes and reads from a lagging replica, they might see things "move backward in time" (e.g., a comment disappears). Solution: Monotonic Reads.

• Ensure a user always reads from the same replica (e.g., hash user ID to replica).

3. Consistent Prefix Reads

In a sharded database, if partitions replicate at different speeds, you might see an answer before the question (causality violation). Solution: Consistent Prefix Reads.

• Writes that are causally related must be written to the same partition.