The correct choice is Adopt a lazy-loading cache pattern because it inserts items into the cache only after a read miss and ensures the cache holds data that clients actually request which helps control memory usage and cost.

With Adopt a lazy-loading cache pattern the application checks Redis on each read and only queries the Aurora MySQL cluster when the key is missing and then populates the cache after the database returns the value. This behavior prevents the cache from being filled with entries that are never requested and it directly reduces the memory footprint and costs for ElastiCache.

Set short TTLs for cached entries is insufficient because TTL only affects how long an item stays after it is cached and it does not stop unneeded keys from being created in the first place which can still waste memory and cause churn.

Switch to a write-through strategy is counter to the goal because it causes the cache to be populated on every database write and that can store items that clients never read which increases cache usage and cost.

Amazon RDS Proxy improves database connection management and pooling but it does not change how or when the cache is populated so it does not meet the requirement to only hold entries after clients request them at least once.

Full AWS Practitioner Certification Question