Tom Lane wrote:
> I wrote:
> > I think it would be a good idea to extend the brinopers table to include
> > the number of expected matches, and to complain if that's not what we got,
> > rather than simply checking for zero.
>
> Also, further experimentation shows that there are about 30 entries in the
> brinopers table that give rise to seqscan plans even when we're commanding
> a bitmap scan, presumably because those operators aren't brin-indexable.
> They're not the problematic cases, but things like
>
> ((charcol)::text > 'A'::text)
>
> Is there a reason to have such things in the table, or is this just a
> thinko? Or is it actually a bug that we're getting such plans?
No, I left those there knowing that there are no plans involving brin --
in a way, they provide some future proofing if some of those operators
are made indexable later.
I couldn't think of a way to test that the plans are actually using the
brin index or not, but if we can do that in some way, that would be
good.
--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services