Do I understand correctly that the author went from a distributed database to a single-master scenario? That's a valid tradeoff, but I'd clearly describe it as such.

My experiences with RethinkDB have been rather positive, but my load is nowhere near that of what the article describes. I agree that ReQL could be improved, I found that there are too many limitations in chaining once you start using it for more complex things.

But the two most important advantages remain for me:

* changefeeds (they work for me),

* a distributed database that I can run on multiple nodes.

I do agree that PostgreSQL is fantastic and that SQL is a fine tool. In my case the above points were the only reasons why I did not use PostgreSQL.

EDIT: after thinking about this for a while, I wonder if the RethinkDB changefeed scenario is doable with the tools in PostgreSQL: get initial user data, then get all subsequent changes to that data, with no race conditions. Many workloads seem to concentrate on twitter-like functionality, where the is no clear concept of a change stream beginning and races do not matter.