I think this is a good answer. Our case is weird; millions of tables with billions (not sure how many in total even) of rows and no joins. Mysql handles this far better in our testing.
May I ask what kind of data you are storing that you need millions of tables? If you have such varied data, what are the tradeoffs compared to only a few tables with json columns?
we’d love to test this scenario with the new OrioleDB storage engine for Postgres.