Skip to content

Multi-region: active-passive and active-active

When a second region is worth it, how active-passive fails over, what active-active costs in capacity and conflicts, and why a cross-region commit costs a round trip.

30 min6 steps

By the end you can
  • Choose between active-passive and active-active for a workload's RPO, RTO and consistency needs
  • Size each region so the survivors can absorb a failed region's traffic
  • Choose how an active-active design handles concurrent writes to the same data
  • Estimate the commit latency of a write that needs a cross-region majority

Active-passive: one region writes

old routeafter DNS TTLasync replicationClientsRegion A (writer)partitioned, maybe aliveRegion B (standby)promoted
Simple (one writer), idle standby. RPO = lag at failure. RTO includes DNS TTLs. Fence the old primary before promoting, or you get two writers (GitHub 2018).

A ledger service needs strongly consistent balances, can tolerate 15 minutes of downtime per regional disaster, and must not lose committed transactions. Which design fits best?

Answer this step to continue.