> You can hand out chunks of sequential ids from a central coordinator to avoid collision; this is a well-established pattern.
The problem is: is that part of postgresql? If not, someone has to write the buggy code for that well-established pattern. (BTW, I honestly think autoincrement is fine and the choice of PK is so minor you can always pay your way to solve it if you really have a problem at scale).
Re: natural keys (or something like it), I was using it as an example of how badly PK choice can impact performance at scale.