I think the point was that saas data is easily sharded as it's all just a bucket of data per user/customer so you have a lot of freedom in distributing customers across different database instances. The most specialized version of that architecture would be an entire DB per customer, but that's likely a lot of potentially unnecessary file overhead (and a giant pain for upgrades and migrations). IMHO you'd probably be fine to just start scaling horizontally distributing users across more DB instances (but really... don't miss the main point that you might not even need more than one DB instance in the first place).
Reports are nice because they can generally be done in an async job that's effectively a big map-reduce run across your DBs. If you need faster real-time reports you're going to want some kind of pipeline for events and stream processing that's outside the scope of this anyways.
Reports are nice because they can generally be done in an async job that's effectively a big map-reduce run across your DBs. If you need faster real-time reports you're going to want some kind of pipeline for events and stream processing that's outside the scope of this anyways.