Route 53 routing policies · Lesson 9 of 10 · 3 min read
Limits of DNS-based traffic management
What DNS can't do
- Instant switches — clients cache answers; TTLs and TTL-ignoring clients slow every change.
- Precise splits — large shared resolvers skew weighted and location decisions.
- Per-request decisions — DNS decides per lookup, not per request; long-lived connections stay where they are.
- Data failover — DNS moves traffic, not data. A regional database failover needs replication and promotion (Aurora Global Database managed failover, DynamoDB global tables, cross-region read replicas promoted to primary) and apps that reconnect.
- Connection draining or retries — that's the load balancer's and client's job.
Enlarge diagram
Swipe sideways to see all of it, or tap to open it full screen.
Complementary AWS services
| Service | Adds |
|---|---|
| Elastic Load Balancing | Per-request balancing, health-based routing, draining, weighted target groups for precise canaries |
| AWS Global Accelerator | Static anycast IPs; regional failover in seconds without DNS caching; traffic dials per region |
| CloudFront | Edge caching, origin failover (origin groups), global TLS termination |
| Route 53 Application Recovery Controller (ARC) | Routing controls (highly available on/off switches for regional traffic, backed by a multi-region cluster), readiness checks (is the standby actually ready — capacity, quotas, config), safety rules |
For a critical API, a senior answer combines: multi-AZ load balancers in each region, Route 53 latency/failover with health checks and low TTLs (or Global Accelerator), ARC routing controls for deliberate, reliable regional evacuation, client retries with backoff, and a tested data-layer failover runbook.