Replicate & Failover

§01Why replication

A single database is a single point of failure: when it goes down, every write and read fails with it. Replication keeps copies of the same data on multiple nodes so reads can keep flowing — and (with promotion) writes too — when one node disappears.

In Flow, click a database in the inspector and press Replicate. The original becomes the primary; new ones are replicas. They share a 🔗 group badge. Reads spread across the group; writes go to the primary.

§02What you get (and what you don't)

• Aggregate read capacity = sum of all healthy members.
• If the primary fails, replicas keep serving reads — your app stays partially up.
• Writes still go to the primary, so during a primary outage they fail. Promotion (a replica becomes the new primary) is a follow-up topic.
• Replication isn't free: every write has to propagate. There's lag.

• Replicate any database that holds important reads — even just one extra copy buys you survivability.
• Replication ≠ free. Lag means a replica might serve a slightly older value.