@alex @lain @graf @n-2-l @mint @Moon Pleroma needs to talk to the DB less. While faster DBs are a nice idea, they will always been significantly slower than retrieving from local process memory. It's not just the overhead of (de)serialization, DBs must be hyper-generic and account for all possible scenarios. The reality is you can make reasonable assumptions about your data and those reasonable assumptions unlock worlds of optimizations.

Reminder best programming talk of all time: https://youtu.be/rX0ItVEVjHc?t=125