Why a local cache in front of Redis, instead of Redis alone?
Because the two layers answer different questions — one removes the network, the other removes the cold start.
Redis alone removes database load. It does not remove the per-lookup network hop, and it makes Redis availability a hard dependency of the request path.
A small, short-TTL in-process cache in front of it removes the hop for the hottest keys and lets the service keep serving — degraded, on slightly stale data — when Redis is unreachable. Redis stays because it is what stops a freshly-scaled instance from stampeding the database to warm itself.
The cost is invalidation across N processes. That cost is real, and it is the reason to only do this for data that is read constantly and written rarely. For anything write-heavy, one level is the right answer.