4.5 Replication Strategies

1. Synchronous Replication

A write is acknowledged only after the required remote replicas have confirmed it. The requirement may be all replicas or a configured quorum; synchronous does not inherently mean every follower.

This reduces the window in which acknowledged data exists on only one node, but write latency and availability now depend on remote replicas.

2. Asynchronous Replication

The leader can acknowledge after committing locally, then replicate to followers later.

Advantages:

• lower write latency;
• a slow follower does not directly block the write path.

Costs:

• follower reads can be stale;
• an acknowledged write may be lost during failover if no surviving replica received it;
• clients may violate read-your-writes, monotonic-read, or prefix expectations unless the system adds stronger routing/version guarantees.

See Distributed Consistency Models.

3. Semi-Synchronous / Quorum Acknowledgement

A leader waits for some, but not necessarily all, replicas before acknowledging. This provides a tunable middle ground between remote durability and latency/availability.

4. Key Point

Replication mode should be described by the exact acknowledgement and failover rules of the product. Labels such as “sync” and “async” are useful shorthand but are not enough to infer consistency guarantees by themselves.