Re: Yet another failure mode in pg_upgrade

On Mon, Sep 3, 2012 at 10:42:38AM -0400, Tom Lane wrote:
> Bruce Momjian <bruce(at)momjian(dot)us> writes:
> > On Mon, Sep 3, 2012 at 10:07:43AM -0400, Tom Lane wrote:
> >> It's not necessary, no? The code now gets socket directory right
> >> without help.
>
> > Well, the doc comment is:
>
> > + If running check on an old pre-9.1 Unix-like running server, and the
> > + old and new servers use different Unix-domain socket directories,
> > + use the <option>-O</> option so the new server uses the same socket
> > + directory as the old server, and set <envar>PGHOST</> similarly.
>
> > Remember, we can't get the socket directory for pre-9.1 servers.
>
> Yeah, but even if you assume that pg_upgrade needs help for that, this
> doc is wrong and overcomplicated. The case that would be problematic
> is where the old server is running with a unix_socket_directory that
> is not the same as the default built into pg_upgrade's libpq. It is
> not necessary (given my patch) that the new server match that socket
> directory. What is necessary is that pg_upgrade be able to contact
> the old server. I think, but haven't tested, that it's sufficient
> to set PGHOST to make that work. The patched code will override
> PGHOST for the new cluster anyway, since it will always specify
> --host as the new-cluster socket directory.

Yes, agreed. With your change, the new server doesn't have to match the
old server's sockdir anymore; it is always going to use the current
directory.

So the text should be:

If running check on an old pre-9.1 Unix-like running server, and
the old running server is using a Unix-domain socket directory
different from the default on the new server, set <envar>PGHOST</>
to point to the socket location of the old server.

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com

+ It's impossible for everything to be true. +