Re: Disk-based hash aggregate's cost model

On Thu, Sep 03, 2020 at 05:53:43PM -0700, Jeff Davis wrote:
>On Tue, 2020-09-01 at 23:19 +0200, Tomas Vondra wrote:
>> FWIW any thoughts about the different in temp size compared to
>> CP_SMALL_TLIST?
>
>Are you referring to results from a while ago? In this thread I don't
>see what you're referring to.
>
>I tried in a simple case on REL_13_STABLE, with and without the
>CP_SMALL_TLIST change, and I saw only a tiny difference. Do you have a
>current case that shows a larger difference?
>

I'm referring to the last charts in the message from July 27, comparing
behavior with CP_SMALL_TLIST fix vs. master (which reverted/replaced the
CP_SMALL_TLIST bit).

Those charts show that the CP_SMALL_TLIST resulted in smaller temp files
(per EXPLAIN ANALYZE the difference is ~25%) and also lower query
durations (also in the ~25% range).

I can repeat those tests, if needed.

[1] https://www.postgresql.org/message-id/20200724012248.y77rpqc73agrsvb3@development

>The only thing I can think of that might change is the size of the null
>bitmap or how fields are aligned.
>

Maybe. Not sure.

regards

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