Ray Ontko <rayo(at)ontko(dot)com> writes:
> Well, that explains the non-determinism, I guess. It seems
> odd that if we aren't doing any updates to the data that analyze
> would give us different stats. How does analyze pick the 3000
> rows it gets back, and should this vary from run to run if we
> aren't changing the table?
At random, and yes. See the documentation ...
regards, tom lane