Running Postgres at the edge of its comfort zone

What happens when you push a single Postgres cluster past 40k queries per second on read replicas. Concrete numbers, the configs that mattered, and the three we reverted.