Opscraft
Route 53 routing policies · Lesson 10 of 10 · 4 min read

Interview questions and model answers

"Name the Route 53 routing policies and when you'd use each."

Simple for a single resource. Weighted to split traffic by percentage — migrations or A/B. Latency to send users to the fastest healthy region in an active-active setup. Failover for active-passive DR with a health check on the primary. Geolocation to route by continent, country or state for compliance or localisation, always with a default record. Geoproximity to route by distance with a bias to shift regional boundaries. Multivalue to return up to eight healthy IPs for client-side spreading. And IP-based to route specific client CIDR ranges, such as an ISP or a customer's offices.

"Latency vs geolocation routing?"

Latency picks the region with the lowest measured latency from the user's network — optimising speed. Geolocation picks based on where the user is — for data residency, licensing or localised content — even if another region would be faster.

"Why not use weighted DNS for a 5% canary?"

Because resolvers cache answers and a few large resolvers serve many users, the actual split is coarse and lumpy, changes slowly, and rollback waits on TTLs. I'd do precise canaries at the ALB with weighted target groups, an ingress controller, or a service mesh, and keep weighted DNS for slow shifts between whole stacks or regions.

"Failover isn't happening although the primary is down. What do you check?"

That the primary record has a health check attached or evaluate-target-health enabled, that the check targets the right endpoint, port and path and actually reports unhealthy, that Route 53 health checker IP ranges are allowed through security groups, that an ALB target truly has no healthy targets, and then the TTL and client caching. I'd also look at the health check's status history in CloudWatch.

"How would you health-check a private endpoint?"

Route 53 health checkers run on the internet, so they can't reach private IPs. I'd create a CloudWatch alarm on a metric that reflects the private endpoint's health — for example a synthetic canary, ALB healthy host count or error rate — and use a CloudWatch alarm-based health check.

"Design global routing: EU users must stay in the EU, everyone else should get the fastest region, with failover."

A geolocation policy at the top: Europe maps to an EU branch, Default maps to a latency branch. The EU branch is a failover pair between two EU regions. The default branch has latency records for each other region with evaluate-target-health or health checks. Nest them with alias records or a Traffic Flow policy, keep TTLs at 60 seconds, and make sure the data layer keeps EU data in EU regions.

"Can you fail over a database with Route 53?"

Route 53 only moves traffic. The database has to be replicated and promoted — Aurora Global Database or a promoted cross-region replica — and the application must reconnect to the new writer. DNS (often the cluster endpoint itself) then points clients to it. Combine with ARC for controlled regional failover and test the runbook regularly.

"Which routing policy have you used and why?" (how to answer)

Name the policy, the business need, the health check and TTL choices, and a lesson learned — for example: "We used weighted records to migrate from the old account's ALB to the new EKS stack, 10% steps over a week, with evaluate-target-health so a bad stack dropped out automatically. TTL was 60. We learned some partner integrations cached DNS for hours, so we kept the old ALB alive for a week after reaching 100%."