Re: Using pg_upgrade on log-shipping standby servers

On Tue, Jul 10, 2012 at 06:21:35PM +0200, Magnus Hagander wrote:
> On Tue, Jul 10, 2012 at 6:17 PM, Bruce Momjian <bruce@momjian.us> wrote:
> > On Tue, Jul 10, 2012 at 12:04:50PM -0400, Tom Lane wrote:
> >> Bruce Momjian <bruce@momjian.us> writes:
> >> > +    While a Log-Shipping Standby Server (<xref linkend="warm-standby">) can
> >> > +    be upgraded, the server must be in changed to a primary server to allow
> >> > +    writes, and after the upgrade it cannot be reused as a standby server.
> >> > +    (Running <command>rsync</> after the upgrade allows reuse.)
> >>
> >> "in changed"?  This sentence makes no sense at all to me.
> >
> > Oops.  New wording attached with "in" removed:
> >
> >         the server must be changed to a primary server
>
> Don't we normally talk about "must be promoted to a primary server",
> not changed?

OK, sure, updated patch attached.

> And wouldn't it be good if it also mentions that another good option
> is to just pg_upgrade the master and rebuild the standby? (Unless
> that's already mentioned somewhere else).

I assume they already realize they re-create the standbys.

> What's the actual usecase for promoting the slave, upgrading it and
> then *not* using it, which is what I think this paragraph suggests?

Testing maybe?  I feel we have just avoided saying what you can and
can't do with the standbys and pg_upgrade, so I think we have to state
something.  If we just want to say "recreate", let's say that.

> And I think the sentence about running rsync is extremely vague - run
> rsync where and how? What are you actually trying to suggest people
> do?

Updated docs attached.

--
  Bruce Momjian  <bruce@momjian.us>        http://momjian.us
  EnterpriseDB                             http://enterprisedb.com

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