In the original article the author suggested on a successful subscription a write to the MailChimp API (because that's the product). In that example, MailChimp is the DB. You don't have to run a copy of a DB locally if you architect things like that. Obviously if the call fails, you need to think about what you fall back to, which is why error handling is such a critical piece of the the Lambda ecosystem - as it should be for every ecosystem, really.