Ray Ontko <rayo(at)ontko(dot)com> writes:
> Does this help?
Well, it says that ANALYZE is not doing anything wrong ;-).
The true null_frac given those numbers is 1.0 - 433809.0/433902.0
or about 0.000214. In the default ANALYZE sample size (3000 rows)
the statistical expectation would be 0.643 NULLs, so most of the
time ANALYZE is going to find 1 NULL, sometimes no NULLs, and rarely
more than 1 NULL. Most of the time you'll get null_frac = 1/3000 =
0.000333333, sometimes 0, etc.
The stats themselves look fine to me; I think there must be a
mathematical error in what selfuncs.c is doing with them.
Still looking to find it...
regards, tom lane