For cache: prioritize availability, low latency, throughput. For durability: enable RDB/AOF on Redis. RDB spikes CPU/memory; AOF reduces write latency but impacts performance. Configure RDB intervals to balance performance/data loss. Avoid AOF every-write mode; everysec is better but can lose data. Use Primary-Replica replication and apply persistence only on replicas to reduce primary load.

Basically, enable Primary-Replica replication using the ReplicaOf command. Make any cache related calls only to the Primary instances. Use the Replicas for persisting cache data. If you end up using AOF (Append-Only File) persistence, use everysec. This way you will continue to get high-throughput, low latency from your primary instances and will have persistence data in case you need to reload with your Replica instances taking on the load of writing the persistence data to disk.