On Sun, Sep 29, 2024 at 08:45:50PM -0400, Thomas Krennwallner wrote:
> if a cluster contains invalid databases that we cannot connect to anymore,
> pg_upgrade would currently fail when trying to connect to the first
> encountered invalid database with
>
> [...]
>
> If there is more than one invalid database, we need to run pg_upgrade more
> than once (unless the user inspects pg_database).
>
> I attached two small patches for PG 17 and PG 18 (can be easily backported
> to all previous versions upon request). Instead of just failing to connect
> with an error, we collect all invalid databases in a report file
> invalid_databases.txt:
Should we have pg_upgrade skip invalid databases? If the only valid action
is to drop them, IMHO it seems unnecessary to require users to manually
drop them before retrying pg_upgrade.
--
nathan
