Authoritative DNS geo-routing failover

Design a global authoritative DNS and traffic-routing system that steers users to the nearest healthy region for a service with 12 regions and 30M queries/sec at peak. It must support geo/latency-based routing, weighted records for gradual region rollout, and automatic failover that pulls a region out of rotation within tens of seconds of it going unhealthy. Constraints: authoritative-answer p99 under 5ms, resolver-side caching means you can't instantly redirect users, and a routing-config change must converge globally fast without serving stale/wrong answers. Walk through the architecture, the health/failover model, and the core trade-off.