I designed and built (with a great team) a very large scale one of these ‘database per resource’ designs just over a decade ago. They have a ton of nice properties (low blast radius, great scalability), but also introduce a lot of complexity (e.g. finding the right database).
This can be a good approach (e.g. see our 2020 paper “Millions of Tiny Databases”), but has some practical problems: durability is hard unless you sync replicate, you need to find the needle in a haystack of the right DB for the right data.