Date: Fri, 29 Sep 2023 11:36:42 -0500
From: Justin Pryzby <pryzby@telsasoft.com>
To: Michael Paquier <michael@paquier.xyz>
Cc: Ali Akbar <the.apaan@gmail.com>,
	Michael Paquier <michael.paquier@gmail.com>,
	Amit Langote <Langote_Amit_f8@lab.ntt.co.jp>,
	tushar <tushar.ahuja@enterprisedb.com>,
	Tom Lane <tgl@sss.pgh.pa.us>, pgsql-hackers@postgresql.org,
	Alvaro Herrera <alvherre@alvh.no-ip.org>
Subject: Re: [HACKERS] pg_upgrade failed with error - ERROR: column "a" in
 child table must be marked NOT NULL
Message-ID: <ZRb9GugoMk74CAn+@pryzbyj2023>
References: <217ddce6-1b0f-509d-f8e0-5ddd9cad5d86@lab.ntt.co.jp>
 <CAB7nPqStji75D1M=DEkpOXsJbF151m=ZLrJ4jXTh-Czk3kKZ1Q@mail.gmail.com>
 <CACQjQLrN88BrpUAFv-soW+bCnDEkkWGc+z6cmSEMfGMpd+Aa=g@mail.gmail.com>
 <107ccc78-1a10-ff59-e837-e5c99dc8c152@lab.ntt.co.jp>
 <CACQjQLoMsE+1pyLe98pi0KvPG2jQQ94LWJ+PTiLgVRK4B=i_jg@mail.gmail.com>
 <20171214054902.GC23527@telsasoft.com>
 <CAB7nPqRgRbMWLaidxKW+n144kT+3xRfTSqNS8OUaV77FuQo48w@mail.gmail.com>
 <CACQjQLqoXTzCV4+-s1XOT62tY8ggkZkH4kY03gm2aQYOMXE+SA@mail.gmail.com>
 <ZRXGAaOvamcRjoL+@telsasoft.com>
 <ZRYXY3WMVep8yEgq@paquier.xyz>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZRYXY3WMVep8yEgq@paquier.xyz>
Archived-At: <https://www.postgresql.org/message-id/ZRb9GugoMk74CAn%2B%40pryzbyj2023>
Precedence: bulk

On Fri, Sep 29, 2023 at 09:16:35AM +0900, Michael Paquier wrote:
> You mean when upgrading from an instance of 9.6 or older as c30f177 is
> not there, right?

No - while upgrading from v15 to v16.  I'm not clear on how we upgraded
*to* v15 without hitting the issue, nor how the "not null" got
dropped...

> Anyway, it seems like the patch from [1] has no need to run this check
> when the old cluster's version is 10 or newer.  And perhaps it should
> mention that this check could be removed from pg_upgrade once v10
> support is out of scope, in the shape of a comment.

You're still thinking of PRIMARY KEY as the only way to hit this, right?
But Ali Akbar already pointed out how to reproduce the problem with DROP
NOT NULL - which still applies to both v16 and v17.