I cannot understand the fuzz about multi-master replication, it either involves dirty/phantom reads, or it is awfully slow. Wikipedia has a lot of examples of these issues https://en.wikipedia.org/wiki/Isolation_%28database_systems%...

There should be a book about the dark side of multi-master replication at scale, about what can you trade away and still be happy with. Preferring either PostgreSQL or MySQL for this is just avoiding the technical complexity that multi-master replication is.