On Tue, Mar 27, 2018 at 11:28 PM, Alvaro Herrera <alvhe...@alvh.no-ip.org> wrote: > Tom Lane wrote: >> Tomas Vondra <tomas.von...@2ndquadrant.com> writes: >> > I think number of index tuples makes sense, as long as that's what the >> > costing needs. That is, it's up to the index AM to define it. But it >> > clearly should not flap like this ... >> >> > And it's not just BRIN. This is what I get with a GIN index: >> >> Sounds like the same kind of thing we just fixed for SP-GiST :-( > > Most likely I modelled the BRIN code after GIN. >
It's better to create a new index AM that estimates the number of index tuples, and to update the index stats using that returned value? Regards, -- Masahiko Sawada NIPPON TELEGRAPH AND TELEPHONE CORPORATION NTT Open Source Software Center