Hacker News Reader

Yes, at this point it's manual optimization task - you have to decide which statistics you need (e.g. by looking for poor estimates in query plans) and then create them. Pretty much just like indexes.

The ultimate goal is to allow detecting this automatically, and either create the statistics right away or at least providing some suggestions to the users. That's not in Pg10 though.

I think the best way to handle this situation is for the default install to be generally performant on any given task, with the option to flip switches for those use cases where extreme performance are needed due to the specific workload.

If that's the case (which I suspect it is, having used prior PSQL versions), then most people will never need to optimize it, but those who have an extreme use case that can benefit from such tweaks will have such options available.

CMU's DBMS is trying to do this: http://pelotondb.io