Re: index v. seqscan for certain values

"Jeremy Dunn" <jdunn@autorevenue.com> writes:
> Interestingly, I tried increasing the stat size for the CID column to
> 2000, analyzing, and checking the accuracy of the stats again.

There's a hard limit of 1000, I believe.  Didn't it give you a warning
saying so?

At 1000 the ANALYZE sample size would be 300000 rows, or about a quarter
of your table.  I would have thought this would give frequency estimates
with much better precision than you seem to be seeing --- but my
statistics are rusty enough that I'm not sure about it.  Possibly the
nonuniform clumping of CID has something to do with the poor results.

Any stats majors on the list?

            regards, tom lane