Hi,
On 2022-02-17 14:27:09 -0500, Robert Haas wrote:
> The other one is trickier, because AFAICT it's basically an opinion
> question: is accessing pg_class in the template database from some backend
> that is connected to another database too ugly to be acceptable? Several
> people have expressed concerns about that, but it's not clear to me whether
> they are essentially saying "that is not what I would do if I were doing
> this project" or more like "if you commit something that does it that way I
> will be enraged and demand an immediate revert and the removal of your
> commit bit." If it's the former, I think it's possible to clean up various
> details of these patches to make them look nicer than they do at present and
> get something committed for PostgreSQL 15.
Could you or Dilip outline how it now works, and what exactly makes it safe
etc (e.g. around locking, invalidation processing, snapshots, xid horizons)?
I just scrolled through the patchset without finding such an explanation, so
it's a bit hard to judge.
> But if it is the latter then there's really no point to that kind of cleanup
> work and we should probably just give up now.
This thread is long. Could you summarize what lead you to consider other
approaches (e.g. looking in the filesystem for relfilenodes) as not feasible /
too ugly / ...?
Greetings,
Andres Freund