Date: Tue, 22 Jun 2021 19:55:06 +0800
From: Julien Rouhaud <rjuju123@gmail.com>
To: "Radoulov, Dimitre" <cichomitiko@gmail.com>
Cc: pgsql-admin@lists.postgresql.org
Subject: Re: Temp files for simple queries
Message-ID: <20210622115506.o2u4u5krplxxc54k@nol>
References: <45cc44dd-b830-5df5-1f09-171e9874bbc0@gmail.com>
 <16aa733d-bcaf-0aa6-f3f0-5206de695e92@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <16aa733d-bcaf-0aa6-f3f0-5206de695e92@gmail.com>
Archived-At: 
 <https://www.postgresql.org/message-id/20210622115506.o2u4u5krplxxc54k%40nol>
Precedence: bulk

On Tue, Jun 22, 2021 at 01:51:55PM +0200, Radoulov, Dimitre wrote:
> Just to clarify: the size of the temp file is 7MB and with work_mem of 4MB
> it clearly doesn't fit.
> 
> But why an "order by" of a 440kB of data would require 7MB.
> 
> One possible reason is that that the number of records in that table varies
> significantly during the day, 

That would be the logical explanation.  You could configure auto_explain to
make sure of that (https://www.postgresql.org/docs/current/auto-explain.html).

> but - if that's the case - I would expect that
> the table would be much bigger than 440kB (size taken using \dt+).

Not necessarily.  autovacuum can truncate the file if all the trailing blocks
are empty (and if it can acquire an exclusive lock fast enough), so if some
client inserts a lot of rows, process them and remove them all, and later on
slowly start to insert new rows you will get that behavior.