When should you cache stuff somewhere other than the database?

hey folks, weve been talking a lot about distributed systems and caching is obviously key. we usually think about caching at the application level or using a dedicated cache like Redis. but what about caching data before it even hits the database for read operations? im thinking about scenarios where a request might be served directly from a cache layer without ever touching our primary data store. are there specific patterns or trade-offs we should consider when implementing this kind of 'pre-database' caching? when does it make sense versus when does it introduce too much complexity or staleness?