Re: Pg upgrade bug with NOT NULL NOT VALID

On Thu, 21 May 2026 at 22:18, Álvaro Herrera <alvherre(at)kurilemu(dot)de> wrote:
>
> Hi,
>

Hi, thank you for looking into this.

> I see two alternatives. One is to have pg_dump --binary-upgrade choose
> a constraint name for the not-null with full knowledge of all other
> constraint names, so that we know to generate a non conflicting one.
> I suspect this is not easy to code.
>

Well, for this option, we need to be told about what other constraint
names that are about to be created. So, pg_dump will need to issue an
SQL that says: please create this relation, but also never choose
name1 to anything in the process. I guess this is not committable...
Maybe you can clarify the design here?

> The other is much simpler: make pg_upgrade -c warn you about the check
> constraint name so that you know to rename it before the upgrade.

I don't think this is good when the database asks you to change your
DDL because of its internal troubles with something.

I also think we should make pg_upgrade just work in this case.

--
Best regards,
Kirill Reshke