practice

Region Strategy

How many regions, where, and in what topology — driven by residency, latency and availability requirements rather than by ambition.

multi-regionlatencyresidencyavailabilitycost

Definition

A region strategy answers: how many regions, which ones, whether they are active-active or active-passive, and where data lives. It is one of the most expensive architectural decisions to change.

The three drivers, and only these

1. Data residency. A legal requirement that data stays in a jurisdiction. This is a hard constraint, not a preference, and it usually forces per-region data isolation with a thin global control plane holding no regulated data.

2. Latency. Users far from the region experience the round trip. Note this is frequently solvable without multi-region infrastructure — a CDN and edge compute address most latency concerns for read-heavy workloads at a fraction of the cost.

3. Availability. Surviving the loss of a region. Ask honestly whether the business requires it: regional failures are rare, and the cost of multi-region is continuous.

If none of these applies, one region with multiple availability zones is the correct answer, and it is the answer for a large majority of systems.

The topologies

Topology Availability Complexity
Single region, multi-AZ Survives zone loss Low — correct default
Active-passive Survives region loss with an RTO Moderate; the standby drifts unless exercised
Active-active, partitioned by user Survives region loss; no write conflicts Moderate — the underrated option
Active-active, shared writes Highest Very high — conflict resolution everywhere

Active-active partitioned by user or tenant is the option most often overlooked. Each user has a home region that owns their writes, so there are no conflicts to resolve, and a failure moves users to another region rather than requiring global consensus. It gives most of the benefit of active-active without the hardest part.

What multi-region actually costs

  • Data consistency. Cross-region replication is asynchronous, so either you accept lag or you pay a round trip on every write.
  • Doubled infrastructure, plus inter-region transfer charges.
  • Operational complexity — deployments per region, drift, and testing.
  • A hard question about the control plane. Identity, secrets, deployment and observability must be multi-region too, or the failover cannot be executed.

Failure scenarios

  • Multi-region adopted for latency where a CDN would have sufficed at a tenth of the cost.
  • A passive region that has drifted, discovered during failover.
  • Active-active with shared writes and last-write-wins conflict resolution, silently losing data.
  • A single-region control plane making the whole strategy inoperable.

Interview question

"A stakeholder asks for multi-region. What three questions do you ask before agreeing?"