Re: Extended statistics is not working on Vars hidden under a RelabelType

On 10/14/2017 07:49 PM, Robert Haas wrote:
> On Fri, Oct 13, 2017 at 4:49 PM, David Rowley
> <david(dot)rowley(at)2ndquadrant(dot)com> wrote:
>> tps = 8282.481310 (including connections establishing)
>> tps = 8282.750821 (excluding connections establishing)
>
> vs.
>
>> tps = 8520.822410 (including connections establishing)
>> tps = 8521.132784 (excluding connections establishing)
>>
>> With the patch we are making use of the extended statistics, which
>> we do expect to be more work for the planner. Although, we didn't
>> add extended statistics to speed up the planner.
>
> Sure, I understand. That's actually a pretty substantial regression
> - I guess that means that it's pretty important to avoid creating
> extended statistics that are not needed, at least for short-running
> queries.
>

Well, it's only about 3% difference in a single run, which may be easily
due to slightly different binary layout, random noise etc. So I wouldn't
call that "substantial regression", at least not based on this one test.

I've done more thorough testing, and what I see is 1.0-1.2% drop, but on
a test that's rather extreme (statistics on empty table). So again,
likely well within noise, and on larger tables it'll get even less
significant.

But of course - it's not free. It's a bit more work we need to do. But
if you don't need multi-column statistics, don't create them. If your
queries are already fast, you probably don't need them at all.

regards

--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services