Database Traffic Control

Deceptive query plans (eg. due to very specific write shapes), lack of QoS classes (to priorize OLTP requests over OLAP ones without spawning a read replica), manual partitioning that is way too manual, lack of backpressure during replication (WAL accumulate and Postgres continues accepting writes) and lack of first-class leader-election mechanism (eg. something like what sorintlab/stolon does) are my top 5 issues with PostgreSQL. Glad to see that one of my 5 pain point (that I call "QoS classes") is shared and addressed, I'm sure on-call engineers will thank you, this work addresses real operating issues.

Hi, I'm the lead engineer for Traffic Control. Glad to see it back in the news! If anyone's interested in more detail about how it works, check out https://planetscale.com/blog/behind-the-scenes-how-traffic-c..., or I'm happy to answer questions here.