Re: hashagg slowdown due to spill changes

On Sat, Jun 13, 2020 at 11:48:09AM -0700, Jeff Davis wrote:
>On Fri, 2020-06-12 at 17:12 -0700, Andres Freund wrote:
>> Do you think we should tackle this for 13? To me 4cad2534da seems
>> like a
>> somewhat independent improvement to spillable hashaggs.
>
>We've gone back and forth on this issue a few times, so let's try to
>get some agreement before we revert 4cad2534da. I added Robert because
>he also seemed to think it was a reasonable idea.
>

I can't speak for Robert, but I haven't expected the extra projection
would be this high. And I agree with Andres it's not very nice we have
to do this even for aggregates with just a handful of groups that don't
need to spill.

In any case, I think we need to address this somehow for v13 - either we
keep the 4cad2534da patch in, or we tweak the cost model to reflect the
extra I/O costs, or we project only when spilling.

I'm not in a position to whip up a patch soon, though :-(


regards

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