Multi-Region Architecture

Why Multi-Region?

Single-region deployments are vulnerable to regional outages, natural disasters, and cloud provider failures. Multi-region architecture distributes your system across geographically distant regions to provide resilience and reduce latency for global users. However, it introduces significant complexity around data consistency and network partitions.

Deployment Models

                    ┌─────────────┐
                    │  Global LB   │
                    │  (Route53)   │
                    └──────┬──────┘
                           │
              ┌────────────┼────────────┐
              │                         │
     ┌────────▼────────┐      ┌────────▼────────┐
     │   US-EAST-1     │      │   EU-WEST-1     │
     │ ┌─────────────┐ │      │ ┌─────────────┐ │
     │ │   App Tier   │◄├──CRDT──┤   App Tier   │ │
     │ └──────┬──────┘ │      │ └──────┬──────┘ │
     │        │        │      │        │        │
     │ ┌──────▼──────┐ │      │ ┌──────▼──────┐ │
     │ │   Database   │ │      │ │   Database   │ │
     │ └─────────────┘ │      │ └─────────────┘ │
     └─────────────────┘      └─────────────────┘

Cross-Region Data Replication

Cross-region replication is the hardest part of multi-region design. The speed of light imposes a minimum latency of 30-100ms between regions. You must choose between synchronous replication (strong consistency, higher latency) and asynchronous replication (lower latency, potential data loss).

1. Async replication — Data written locally, replicated asynchronously. Low latency, RPO > 0.
2. Synchronous replication — Data written to all regions before acknowledging. Strong consistency, high latency.
3. Semi-sync — Write to local + one remote. Balance between latency and durability.
4. CRDT-based — Conflict-free data types that can merge without coordination. Best for active-active.

Latency Tradeoffs

Every cross-region operation adds 30-100ms of network latency. Design your system to minimize cross-region calls in the hot path. Read from the local region, write asynchronously, and use regional data stores where possible.

interface RegionConfig {
  region: string;
  primary: boolean;
  endpoints: string[];
  replicationLagMs: number;
}

function routeRequest(userRegion: string): RegionConfig {
  const region = getNearestRegion(userRegion);
  
  // If nearest region is healthy, use it
  if (isHealthy(region)) return region;
  
  // Failover to another region
  return regions.find(r => r.primary && isHealthy(r))!;
}

Routing Strategies

Global routing determines which region handles each request. The strategy affects latency, cost, and consistency. Common approaches include geo-DNS (Route53 latency-based), anycast (BGP), and application-level routing.

Disaster Recovery Runbooks

A multi-region architecture without a tested disaster recovery runbook is just expensive redundancy. Document the exact steps for regional failover, data recovery, and failback. Run quarterly DR drills.