Re: Multivariate MCV list vs. statistics target

On Tue, 18 Jun 2019 at 22:34, Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com> wrote:
>
> One slightly inconvenient thing I realized while playing with the
> address data set is that it's somewhat difficult to set the desired size
> of the multi-column MCV list.
>
> At the moment, we simply use the maximum statistic target for attributes
> the MCV list is built on. But that does not allow keeping default size
> for per-column stats, and only increase size of multi-column MCV lists.
>
> So I'm thinking we should allow tweaking the statistics for extended
> stats, and serialize it in the pg_statistic_ext catalog. Any opinions
> why that would be a bad idea?
>

Seems reasonable to me. This might not be the only option we'll ever
want to add though, so perhaps a "stxoptions text[]" column along the
lines of a relation's reloptions would be the way to go.

> I suppose it should be part of the CREATE STATISTICS command, but I'm
> not sure what'd be the best syntax. We might also have something more
> similar to ALTER COLUMNT, but perhaps
>
> ALTER STATISTICS s SET STATISTICS 1000;
>
> looks a bit too weird.
>

Yes it does look a bit weird, but that's the natural generalisation of
what we have for per-column statistics, so it's probably preferable to
do that rather than invent some other syntax that wouldn't be so
consistent.

Regards,
Dean