Feedback-ID: ia2694551:Fastmail
Date: Tue, 22 Oct 2024 16:32:33 +0200
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Jehan-Guillaume de Rorthais <jgdr@dalibo.com>
Cc: Tender Wang <tndrwang@gmail.com>, Junwang Zhao <zhjwpku@gmail.com>,
	"pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>,
 Alexander Lakhin <exclusion@gmail.com>,
	Guillaume Lelarge <guillaume@lelarge.info>,
 Michael Paquier <michael@paquier.xyz>
Subject: Re: [BUG] Fix DETACH with FK pointing to a partitioned table fails
Message-ID: <202410221432.3qtqbk2q6o7q@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <20241022162312.63021cc6@karst>
Archived-At: 
 <https://www.postgresql.org/message-id/202410221432.3qtqbk2q6o7q%40alvherre.pgsql>
Precedence: bulk

On 2024-Oct-22, Jehan-Guillaume de Rorthais wrote:

> On Fri, 18 Oct 2024 16:50:59 +0200
> Alvaro Herrera <alvherre@alvh.no-ip.org> wrote:

> > For branch 14, I opted to make it delete the constraints on detach.
> > This isn't ideal but unless somebody wants to spend a lot more time on
> > this, it seems the best we can do.  Leaving broken constraints around
> > seems worse. 
> 
> Keep that in mind, and move ahead to the next level: self-FK on partitioned
> table! ;-)
> 
> https://www.postgresql.org/message-id/flat/20230707175859.17c91538%40karst#0dc7b8afd8b780899021bbb075598250

Yeah.  I pushed these patches finally, thanks!

Now we can discuss what to do about self-referencing FKs ... maybe we
should consider dropping the FK on detach in problematic cases.

> > However, in the end I think this is a very good technique to verify that
> > the fix works correctly, but it's excessive to include these results in
> > the tests forevermore.  So I've left them out for now.  Maybe we should
> > reconsider on the older branches?
> 
> The point here was to make sure futur work/refactoring don't forget/break
> anything in the catalog representation of FK on partitions.

There is that ... but I think testing for user-visible symptoms is
better.  I'm not sure I want to bet against the odds that this will
become make-work to keep the output of internal catalog state up to date
for whatever other changes we need to do.

-- 
Álvaro Herrera        Breisgau, Deutschland  —  https://www.EnterpriseDB.com/
 Are you not unsure you want to delete Firefox?
       [Not unsure]     [Not not unsure]    [Cancel]
                   http://smylers.hates-software.com/2008/01/03/566e45b2.html