Not to mention the challenges with multi-region replication needs... to do this for every request along a server-server pipeline adds more latency still, since each request to the db means potentially 2-3ms on top of more complex requests, which all adds up.