On Mon, Feb 22, 2016 at 4:32 AM, Tomas Vondra
<tomas.vondra@2ndquadrant.com> wrote:
> Firstly, we could clamp the join cardinality estimate to the size of
> cartesian product, eliminating at least the most obvious over-estimates.
>
> Secondly, I think it'd be nice to somehow eliminate the sudden jumps in the
> estimates - I'm not quite sure why need the three different formulas in
> eqjoinsel_semi, so maybe we can get rid of some of them?
I don't have any particularly intelligent commentary on the second
point right at the moment, but I emphatically agree with you on the
first one. I've seen that problem a number of times and it seems
completely ludicrous to me.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company