Re: Minmax indexes

On 2014-06-17 16:48:07 -0700, Greg Stark wrote:
> On Tue, Jun 17, 2014 at 11:16 AM, Andres Freund <andres(at)2ndquadrant(dot)com> wrote:
> > Well, it might be doable to correlate them along one axis, but along
> > both? That's more complicated... And even alongside one axis you
> > already get into problems if your geometries are irregularly sized.
> > Asingle large polygon will completely destroy indexability for anything
> > stored physically close by because suddently the minmax range will be
> > huge... So you'll need to cleverly sort for that as well.
>
> I think there's a misunderstanding here, possibly mine. My
> understanding is that a min/max index will always be exactly the same
> size for a given size table. It stores the minimum and maximum value
> of the key for each page. Then you can do a bitmap scan by comparing
> the search key with each page's minimum and maximum to see if that
> page needs to be included in the scan. The failure mode is not that
> the index is large but that a page that has an outlier will be
> included in every scan as a false positive incurring an extra iop.

I just rechecked, and no, it doesn't, by default, store a range for each
page. It's MINMAX_DEFAULT_PAGES_PER_RANGE=128 pages by
default... Haven't checked what's the lowest it can be se tto.

Greetings,

Andres Freund

--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services