We used PostgreSQL before and while excellent on a single node it's deeply painful to cluster. You need a full time DBA to make sure you're doing it right, as GitLab discovered, and when it fails it gives error messages that seem more cryptic than C++ template errors. Amazon RDS was insanely pricey compared to bare metal (our load is both CPU-bound and disk-IO-heavy bare metal on OVH is >10X cheaper than Amazon) and Redshift didn't have pgsql syntax yet. Even now that it does I'm not sure I want to be bound to it as I have an instinctive aversion to cloud lock-in.
I think the big thing for us is that everything about RethinkDB screams quality from a technical POV. It's powerful but not over-engineered. The code is pretty clean looking. It "just works" without hiding wads of chewing gum and duct tape in Docker containers or requiring 10-step setup processes. When you do bang on it ChaosMonkey-style it holds up very well.