Part 2 of 5 · 1 min read
Why more than one leader
A single leader forces every write across long distances to one place; multiple leaders let writes happen locally instead.
Think of a company with offices on opposite sides of the world. Every single order form currently has to be physically mailed to one central headquarters before it can be approved, adding real delay to every single order regardless of which office it came from.
In plain words: single-leader replication, covered two pages ago, works well when all writes can reasonably reach one leader quickly. When users are spread across distant locations, routing every write to one single, far-away leader adds real latency to every write, regardless of where the user actually is. Multi-leader replication addresses this by allowing more than one leader, often one per region, each accepting writes locally and replicating those changes to the other leaders, letting a write happen close to the user who made it.
A worked example. A company with offices in two distant regions, each running its own leader, lets each region's writes commit locally and quickly, with changes replicated to the other region's leader afterward rather than before.
The key lines. No new SQL; which leader a write is routed to is a configuration and routing decision, not something a single statement controls.
Multi-leader replication
Multi-leader replication allows more than one leader to accept writes, typically one per region, letting writes commit locally and quickly, with changes replicated to the other leaders afterward.
Example: each distant regional office approving its own orders locally, instead of mailing every order to one central headquarters, mirrors multi-leader replication's local write acceptance.