Scalability and high availability: multi-AZ and active-active topology, global load balancing and failover, HPA and KEDA autoscaling on custom or queue-depth metrics, circuit breakers, bulkheads, retries with jitter, and rate limiting. Use when a service must survive an availability zone failure, when it collapses under traffic spikes and drags downstream services with it, or when CPU is the wrong autoscaling signal.
