"Clinton James" <cjames@callone.net> writes:
> True, using a different cust_num, it is possible there may be a couple of
> cust_num where that would be true. The majority will not have even near
> that many. In this case there are only 13 matching records. Is VACUUMDB
> ANALYZE using the worst case (the cust_num with the greatest number of rows)
> senario for this?
Yes, because that's as far as the statistics go at the moment :-(.
The frequency of the most common value is basically all the info the
planner has about the data distribution, and so a highly skewed
distribution leads to bad plan choices.
I am presently working on better statistics ... to appear in 7.2 ...
regards, tom lane