Date: Mon, 20 Dec 2021 22:51:38 -0600
From: Justin Pryzby <pryzby@telsasoft.com>
To: Lars Bergeson <larsavatar@gmail.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, pgsql-performance@lists.postgresql.org
Subject: Re: Query is slower with a large proportion of NULLs in several
 columns
Message-ID: <20211221045138.GZ17618@telsasoft.com>
References: 
 <CAK66RGYim22Sd8-NvmufcxEQ_P99=UTAyJ9agnwDyEbCiVGEQg@mail.gmail.com>
 <2688019.1640051519@sss.pgh.pa.us>
 <CAK66RGbd8n46xEp3fTKOa3pO3eC=NT8tEHHZYjuNZq4Y0SQHTw@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: 
 <CAK66RGbd8n46xEp3fTKOa3pO3eC=NT8tEHHZYjuNZq4Y0SQHTw@mail.gmail.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Archived-At: 
 <https://www.postgresql.org/message-id/20211221045138.GZ17618%40telsasoft.com>
Precedence: bulk

On Mon, Dec 20, 2021 at 08:11:42PM -0800, Lars Bergeson wrote:
> ok, here are results after I did:
> set max_parallel_workers_per_gather = 0;
> 
> HashAggregate  (cost=1676432.13..1676432.16 rows=3 width=15) (actual time=19908.343..19908.345 rows=5 loops=1)
>   I/O Timings: read=532369.898
> Execution Time: 19908.383 ms

> HashAggregate  (cost=1390580.70..1390580.72 rows=2 width=15) (actual time=30369.758..30369.761 rows=5 loops=1)
>   I/O Timings: read=6440851.540
> Execution Time: 30369.796 ms

> Still taking 10X more I/O to read the smaller table. Very odd.

If I'm not wrong, it's even worse than that ?
It takes 20 or 30sec to run the query - but it says the associated I/O times
are ~500sec or ~6000sec ?

What architecture and OS/version are you running ?
How did you install postgres?  From a package or compiled from source ?

It might be interesting to know the output from something like this command,
depending on whether and where the headers like pg_config_x86_64.h are installed.

grep -r HAVE_CLOCK_GETTIME /usr/pgsql-13/include

-- 
Justin