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.