Date: Mon, 30 Nov 2020 21:03:45 -0300
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Cc: keisuke.kuroda.3862@gmail.com, tatsuro.yamada.tf@nttcom.co.jp,
	pgsql-hackers@lists.postgresql.org, amitlangote09@gmail.com,
	tatsuhito.kasahara.rd@hco.ntt.co.jp
Subject: Re: Huge memory consumption on partitioned table with FKs
Message-ID: <20201201000345.GA15098@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20201126.121818.26523414172308697.horikyota.ntt@gmail.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On 2020-Nov-26, Kyotaro Horiguchi wrote:

> This shares RI_ConstraintInfo cache by constraints that shares the
> same parent constraints. But you forgot that the cache contains some
> members that can differ among partitions.
> 
> Consider the case of attaching a partition that have experienced a
> column deletion.

I think this can be solved easily in the patch, by having
ri_BuildQueryKey() compare the parent's fk_attnums to the parent; if
they are equal then use the parent's constaint_id, otherwise use the
child constraint.  That way, the cache entry is reused in the common
case where they are identical.

I would embed all this knowledge in ri_BuildQueryKey though, without
adding the new function ri_GetParentConstOid.  I don't think that
function meaningful abstraction value, and instead it would make what I
suggest more difficult.