Re: Bloom index cost model seems to be wrong

On Tue, Feb 12, 2019 at 4:17 PM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> Jeff Janes <jeff(dot)janes(at)gmail(dot)com> writes:
> > In order for bloom (or any other users of CREATE ACCESS METHOD, if there
> > are any) to have a fighting chance to do better, I think many of
> selfuncs.c
> > currently private functions would have to be declared in some header
> file,
> > perhaps utils/selfuncs.h. But that then requires a cascade of other
> > inclusions. Perhaps that is why it was not done.
>
> I'm just in the midst of refactoring that stuff, so if you have
> suggestions, let's hear 'em.
>

The goal would be that I can copy the entire definition of
genericcostestimate into blcost.c, change the function's name, and get it
to compile. I don't know the correct way accomplish that. Maybe
utils/selfuncs.h can be expanded to work, or if there should be a new
header file like "utils/index_am_cost.h"

What I've done for now is:

#include "../../src/backend/utils/adt/selfuncs.c"

which I assume is not acceptable as a real solution.

It's possible that a good cost model for bloom is so far outside
> genericcostestimate's ideas that trying to use it is not a good
> idea anyway.
>

I think that might be the case. I don't know what the right answer would
look like, but I think it will likely end up needing to access everything
that genericcostestimate currently needs to access. Or if bloom doesn't,
some other extension implementing an ACCESS METHOD will.

Cheers,

Jeff