Subject: Re: Postgres chooses slow query plan from time to time
To: Kristjan Mustkivi <sonicmonkey@gmail.com>,
 pgsql-performance@lists.postgresql.org
References: 
 <CAOQPKavhdTrgQo8Uss0M+K0Du+_WYmdoGyMpC0hJQkKM+-X25Q@mail.gmail.com>
From: Tomas Vondra <tomas.vondra@enterprisedb.com>
Message-ID: <31c01752-c4d6-7751-9d25-b1fac8f3a7aa@enterprisedb.com>
Date: Mon, 13 Sep 2021 15:50:46 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.11.0
MIME-Version: 1.0
In-Reply-To: 
 <CAOQPKavhdTrgQo8Uss0M+K0Du+_WYmdoGyMpC0hJQkKM+-X25Q@mail.gmail.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Archived-At: 
 <https://www.postgresql.org/message-id/31c01752-c4d6-7751-9d25-b1fac8f3a7aa%40enterprisedb.com>
Precedence: bulk

On 9/13/21 3:24 PM, Kristjan Mustkivi wrote:
> Dear community,
> 
> I have a query that most of the time gets executed in a few
> milliseconds yet occasionally takes ~20+ seconds. The difference, as
> far as I am able to tell, comes whether it uses the table Primary Key
> (fast) or an additional index with smaller size. The table in question
> is INSERT ONLY - no updates or deletes done there.
> 

It'd be really useful to have explain analyze for the slow execution.

My guess is there's a poor estimate, affecting some of the parameter
values, and it probably resolves itself after autoanalyze run.

I see you mentioned SET STATISTICS, so you tried increasing the
statistics target for some of the columns? Have you tried lowering
autovacuum_analyze_scale_factor to make autoanalyze more frequent?

It's also possible most values are independent, but some values have a
rather strong dependency, skewing the estimates. The MCV would help with
that, but those are in PG12 :-(


regards

-- 
Tomas Vondra
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company