An application interacts with Amazon Aurora to store and track customer information. The primary database is set up with multiple read replicas for improving the performance of the read queries. Howev

Aurora's reader endpoint uses DNS-based load balancing to distribute connections across available read replicas on each new DNS resolution. When applications cache DNS responses, they keep connecting to the same resolved IP (one replica) instead of allowing the DNS TTL to expire and re-resolve to a different replica. Disabling DNS caching forces fresh DNS lookups and restores even traffic distribution.

Enabling DNS caching is the root cause of the problem; it pins connections to a single replica and prevents load balancing from working.

Connection pooling manages the reuse of existing database connections but does not affect how DNS resolution distributes traffic across replicas.

Disabling connection pooling would increase connection overhead and does not address the DNS-based load balancing issue.