An application uses Amazon Aurora with multiple read replicas to scale read queries. One replica is receiving most or all traffic while another sits idle. How can you fix this imbalance?

Disabling application-level DNS caching is the correct solution. Aurora read replicas use a single endpoint for all replicas, and DNS resolution rotates among the available replica IP addresses. If the application caches DNS resolutions, it will continue sending traffic to the same IP address even as other replicas become available or the load shifts. This prevents effective load balancing across the replicas. Enabling application-level DNS caching would exacerbate the problem. Connection pooling manages database connections but doesn't directly address DNS-based load balancing across read replicas.