Re: Official Windows Installer and Documentation

On Wed, Jul 27, 2022 at 8:22 PM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> I wrote:
> > If EDB isn't adequately filling in the documentation for the behavior
> > of their packaging, that's on them.
>
> Having now looked more closely at the pg_upgrade documentation,
> I don't think this is exactly EDB's fault; it's text that should
> never have been there to begin with. ISTM we need to simply rip out
> lines 431..448 of pgupgrade.sgml, that is all the Windows-specific
> text starting with
>
> For Windows users, you must be logged into an administrative account,
> and
>
> That has got nothing to recommend it: we do not generally provide
> platform-specific details in these man pages, and to the extent it
> provides details, those details are likely to be wrong.

I mean, we do provide platform-specific details/examples, it's just that
platform is a source installed Linux platform (though pathless)

Does the avoidance of dealing with other platforms also apply to NET STOP
or do you find that an acceptable variance? Or are you suggesting that
basically all O/S commands should be zapped? If not, then rewriting 442 to
446 to just be the command seems worthwhile. I'd say pg_upgrade warrants
an examples section like pg_basebackup has (though obviously pg_upgrade is
procedural).

I do have another observation:

https://github.com/postgres/postgres/blob/4fc6b6eefcf98f79211bb790ee890ebcb05c178d/src/bin/pg_upgrade/check.c#L665

if (PQntuples(res) != 1 ||
atooid(PQgetvalue(res, 0, 1)) != BOOTSTRAP_SUPERUSERID)
pg_fatal("database user \"%s\" is not the install user",
os_info.user);

Any reason to not inform the DBA the name of the install user here? Sure,
it is almost certainly postgres, but it also seems like an easy win in
order for them, and anyone they may ask for help, to know exactly the name
of install user in the clusters should that end up being the issue.
Additionally, from what I can tell, if that check does fail (or any of the
checks really) it is not possible to tell whether the check was being
performed against the old or new server. The user does not know that
checks against the old server are performed first then checks against the
new one, and there are no banners saying "checking old/new"

David J.