Scaling ElastiCache

Correct Answer: Read replicas

Read replicas allow horizontal scaling of read traffic in ElastiCache for Redis. Sharding is used for partitioning data among cluster nodes, while multi-AZ is for high availability. Vertical scaling involves resizing the node, and Global Datastore is for disaster recovery. A Cache Cluster is simply a collection of cache nodes in a region.

Correct Answer: Increase the maximum number of connections

When ElastiCache for Redis experiences high latency while CurrConnections approaches the maxclients limit, client requests are queued or refused, increasing response times. Creating and applying a custom parameter group and increasing the maximum number of connections (Redis maxclients) allows more concurrent clients to be served, reducing queuing and latency, provided the node has sufficient CPU and memory headroom. Enabling the slow query log is diagnostic and does not directly reduce latency. Decreasing the cache hit ratio would worsen performance. "Write coalescing" is not a standard ElastiCache configuration and would not address connection saturation.

Correct Answer: Add more read replicas

Adding more read replicas is the correct way to scale read throughput horizontally for an ElastiCache for Redis deployment, reducing read latency during short, read-heavy bursts such as periodic leaderboard fetches. Each replica can serve read requests, distributing load and lowering latency without changing node families, enabling cluster mode, or using cross-Region features. Online resharding applies only to Redis Cluster mode when changing shard count, which is not applicable here. Setting up a cache cluster is ambiguous and not a specific performance optimization. Global datastore provides cross-Region replication and does not improve in-Region read latency.