Re: New design for FK-based join selectivity estimation

Tomas Vondra <tomas(dot)vondra(at)2ndquadrant(dot)com> writes:
> One of the recent issues with the current design was handling of
> inheritance / appendrels. ISTM the proposed design has the same issue,
> no? What happens if the relations are partitioned?

I haven't thought about inheritance in this proposal. My initial feeling
is that considering the parent table's outgoing FKs (if any) as valid is
not unreasonable. If it has any, probably all the children do too.
Not sure about incoming FKs, but there probably are none anyhow, since
our implementation doesn't really permit reasonable FK definitions that
reference a partitioned table. In any case, whatever we might choose
to do differently for inheritance would be no harder in this scheme than
what's there now; plus, whatever it is, we'd do it once not once per join
relation.

regards, tom lane