Re: do only critical work during single-user vacuum?

On Wed, Feb 16, 2022 at 3:21 PM Peter Geoghegan <pg(at)bowt(dot)ie> wrote:
> On Wed, Feb 16, 2022 at 12:11 PM Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> > No, I think it's PostgreSQL 13, because before the vacuum failsafe
> > thing you could end up truncating enough tables during vacuum
> > operations to actually wrap around.
>
> Why wouldn't the xidStopLimit thing prevent actual incorrect answers
> to queries, even on Postgres 13? Why wouldn't that be enough, even if
> we make the most pessimistic possible assumptions?
>
> To me it looks like it's physically impossible to advance an XID past
> xidStopLimit, unless you're in single user mode. Does your concern
> have something to do with the actual xidStopLimit value in shared
> memory not being sufficiently protective in practice?

No, what I'm saying is that people running older versions routinely
run VACUUM in single-user mode because otherwise it fails due to the
truncation issue. But once they go into single-user mode they lose
protection.

> > And even in 14+, you can still do that, if you use single user mode.
>
> So what you're saying is that there is *some* reason for vacuuming in
> single user mode after all, and so we should keep the advice about
> that in place? :-)

We could perhaps amend the text slightly, e.g. "This is a great idea
if you like pain."

--
Robert Haas
EDB: http://www.enterprisedb.com