Opscraft
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 diagramInstant switch for everyoneGlobal Accelerator (anycast IPs) or the load balancerPrecise percentagesALB weighted target groups, service meshPer-request decisionsL7 load balancer or CDN rulesKnowing about server loadload balancer health and least-outstanding routingDNS cannot…use instead
Figure 1. What DNS can and cannot do for traffic management, and where to look instead.
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.