Re: Bitmap index scans use of filters on available columns

On Fri, Nov 6, 2015 at 10:28 AM, Kyotaro HORIGUCHI <
horiguchi(dot)kyotaro(at)lab(dot)ntt(dot)co(dot)jp> wrote:
>
> Hello,
>
> At Fri, 6 Nov 2015 09:49:30 +0530, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com>
wrote in <CAA4eK1L_5T-UYsQeMOrX54g3QeXGhhAk5YFmzZqu5MidzxzkRg(at)mail(dot)gmail(dot)com
>
> > Apart from other problems discussed, I think it could also lead to
> > a performance penality for the cases when the qual condition is
> > costly as evaluating such a qual against lot of dead rows would be a
> > time consuming operation. I am not sure, but I think some of the
> > other well know databases might not have any such problems as
> > they store visibility info inside index due to which they don't need to
> > fetch the heap tuple for evaluating such quals.
>
> I was junst thinking of the same thing. Can we estimate the
> degree of the expected penalty using heap statistics? Of couse
> not so accurate though.
>

I think so. Information related to number of tuples inserted, updated,
hot-updated, deleted and so on is maintained and is shown by
pg_stat_all_tables. By the way, whats your point here, do you mean to
say that we can estimate approximate penality and then decide whether
quals should be evaluated at index level?

With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com