Re: pg_upgrade problem with invalid indexes

On Fri, Dec 7, 2012 at 03:24:46PM +0100, Andres Freund wrote:
> On 2012-12-07 09:21:58 -0500, Bruce Momjian wrote:
> > On Thu, Dec 6, 2012 at 10:27:21PM -0500, Stephen Frost wrote:
> > > * Bruce Momjian (bruce(at)momjian(dot)us) wrote:
> > > > On Thu, Dec 6, 2012 at 09:45:11PM -0500, Stephen Frost wrote:
> > > > > Or preserve it as-is. I don't really like the 'make them fix it'
> > > > > option, as a user could run into that in the middle of a planned upgrade
> > > > > that had been tested and never had that come up.
> > > >
> > > > They would get the warning during pg_upgrade --check, of course.
> > >
> > > Sure, if they happened to have a concurrent index creation going when
> > > they ran the check... But what if they didn't and it only happened to
> > > happen during the actual pg_upgrade? I'm still not thrilled with this
> > > idea of making the user have to abort in the middle to address something
> > > that, really, isn't a big deal to just preserve and deal with later...
> >
> > If a concurrent index creation was happening during the check,
> > pg_upgrade --check would fail. I don't think there is any indication if
> > the index is failed, or in process.
>
> There should be a lock on the table + index if the creation is in
> progress.

Well, it is a CONCURRENT index creation, so locking would be minimal.
Do we want pg_upgrade to be groveling over the lock view to look for
locks? I don't think so.

> > That is a good argument for _not_ throwing an error because index
> > creation is more of an intermediate state.
>
> Uhm. If pg_upgrade is actually running its definitely not an
> intermediate state anymore...

It would be running pg_upgrade --check, which can be run while the old
server is running.

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

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