Re: So, is COUNT(*) fast now?

"Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov> writes:
> Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> I had wondered whether it'd be worth optimizing that along the
>> lines of slot_getallattrs(). But most indexes probably have only
>> one column, or anyway not enough to make for a useful savings.

>> From a heavily-used production database:

> cir=> select indnatts, count(*) from pg_index group by indnatts
> order by indnatts;
> indnatts | count
> ----------+-------
> 1 | 200
> 2 | 684
> 3 | 155
> 4 | 76
> 5 | 43
> 6 | 13
> 7 | 2
> 9 | 1
> (8 rows)

> This includes system table and TOAST table indexes (which seem to
> have two columns).

Yeah, TOAST indexes are 2-column. It would be best to exclude those
from your counts, since it seems pretty unlikely that anyone will care
how fast nodeIndexonlyscan.c is for scans on toast tables.

> There are over 400 user tables, each of which
> has a primary key, so most primary keys in our database are more
> than one column.

It doesn't look to me like the mean is above 2 (unless you have many
fewer toast tables than I suspect), so trying to optimize many-column
cases isn't going to help.

regards, tom lane