Date: Tue, 17 Nov 2020 08:20:24 -0600
From: Justin Pryzby <pryzby@telsasoft.com>
To: Frits Jalvingh <jal@etc.to>
Cc: pgsql-performance@postgresql.org
Subject: Re: Postgres using nested loops despite setting enable_nestloop to
 false
Message-ID: <20201117142024.GA24784@telsasoft.com>
References: 
 <CAKhTGFXJh6QfZTTXw4Qb-7oiFwgqh7Pdz0nUKjZnayuZpwjGqg@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: 
 <CAKhTGFXJh6QfZTTXw4Qb-7oiFwgqh7Pdz0nUKjZnayuZpwjGqg@mail.gmail.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Precedence: bulk

On Tue, Nov 17, 2020 at 02:47:55PM +0100, Frits Jalvingh wrote:
> But lately while migrating to Postgres 13 (from 9.6) we found that Postgres
> does not (always) obey the enable_nestloop = false setting anymore: some
> 
> The execution plan on Postgres 13.1:

Could you send the plans under pg13 and pg9.6 as attachments ?

What is the setting of work_mem ?

I see the cost is dominated by 2*disable_cost, but I wonder whether the I/O
cost of hash joins now exceeds that.  Maybe hash_mem_multiplier helps you?

GroupAggregate  (cost=20008853763.07..20008853776.02 rows=370 width=68)                                                                                                                                                            
  Group Key: (COALESCE(adres_pe.id_s, '-1'::integer)),                                                                                                                                                                             

-- 
Justin