Bruce Momjian wrote:
> Lou Picciano wrote:
> > After running an essentially uneventful* pg_upgrade from 9.0.4 ->
> > 9.1rc1, we are seeing some toast errors logged on the new cluster:
> >
> > All are of this pattern: ERROR: missing chunk number 0 for toast value
> > 130087 in pg_toast_34735
> >
> > Have seen the same pattern for a few of the databases in the 9.1rc1
> > cluster, and all as a result of a select on a usr table (the offending
> > SELECT happens to be the first one any of these DBs sees, as it's the
> > first step in a user authentication process). SELECT count(*) does not
> > produce an error.
> >
> > *almost uneventful: We also saw messages that the destination cluster
> > did not have one of our schema - (of course it didn't!) - I didn't
> > realize pg_upgrade doesn't 'do' schema?
> I have fixed two errors in pg_upgrade since 9.1rc1 was released. Can
> you use git 9.1 head or wait for 9.1rc2 or 9.1 final? I can email you
> patches if you prefer.
Thinking some more, none of these errors was fixed by the patches I
applied.
The schema error seems very odd --- pg_upgrade certainly handles
schemas. In fact, any error makes pg_upgrade stop, so I am curious what
the error was. Did the upgrade fail and you just started the new
server? That isn't good.
-- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB
http://enterprisedb.com
+ It's impossible for everything to be true. +
