I got jumped on for the “no connections” thing. It’s more about instantiating connections and understanding how functions scale (concurrency, cold starts) and then imagining what would happen under different load scenarios.
You *can* store some state inside a Lambda 😲 and use a cache (I have seen it!) but most of the time it’s best to store state in a data store that is very quick to connect/disconnect on-demand and can rapidly scale e.g. DynamoDB. This usually means it’s not an RDBMS (or at least, not the way it’s usually used).