Re: Additional improvements to extended statistics

On Fri, 6 Mar 2020 at 12:58, Tomas Vondra <tomas.vondra@2ndquadrant.com> wrote:
>
> Here is a rebased version of this patch series. I've polished the first
> two parts a bit - estimation of OR clauses and (Var op Var) clauses.
>

Hi,

I've been looking over the first patch (OR list support). It mostly
looks reasonable to me, except there's a problem with the way
statext_mcv_clauselist_selectivity() combines multiple stat_sel values
into the final result -- in the OR case, it needs to start with sel =
0, and then apply the OR formula to factor in each new estimate. I.e.,
this isn't right for an OR list:

        /* Factor the estimate from this MCV to the oveall estimate. */
        sel *= stat_sel;

(Oh and there's a typo in that comment: s/oveall/overall/).

For example, with the regression test data, this isn't estimated well:

  SELECT * FROM mcv_lists_multi WHERE a = 0 OR b = 0 OR c = 0 OR d = 0;

Similarly, if no extended stats can be applied it needs to return 0
not 1, for example this query on the test data:

  SELECT * FROM mcv_lists WHERE a = 1 OR a = 2 OR d IS NOT NULL;

It might also be worth adding a couple more regression test cases like these.

Regards,
Dean