This explains it all.
What would be involved in adding version and visibility to the index?
It would allow for scanning the index instead of the whole table for
many of the count() request.
JLL
Neil Conway wrote:
>
> Jean-Luc Lachance <jllachan@nsd.ca> writes:
> > unless id is indexed there is nothing that can be done with
> > select count(*) from table where id >10000;
> > Otherwise, the index should be scanned, not the table.
>
> Indexes don't store heap tuple visibility information; you'd need to
> scan the heap as well in order to determine which tuples your
> transaction can see.
>
> Cheers,
>
> Neil
>
> --
> Neil Conway <neilc@samurai.com> || PGP Key ID: DB3C29FC