> > -----Original Message-----
> > From: pgsql-hackers-owner@hub.org [mailto:pgsql-hackers-owner@hub.org]On
> > Behalf Of Tom Lane
> >
> > Tim Perdue <tperdue@valinux.com> writes:
> > >>>>>> -now hopefully I can create my indexes and be back in business
> >
> > > I vacuumed here and it worked. I did not use my "old" pg_log
> file - what
> > > did I lose?
> >
> > Hard to tell. Any tuples that weren't already marked on disk as "known
> > committed" have probably gone missing, because their originating
> > transaction IDs likely won't be shown as committed in the new pg_log.
> > So I'd look for missing tuples from recent transactions in the old DB.
> >
>
> Hmm,this may be more serious.
> MVCC doesn't see committed(marked HEAP_XMIN_COMMITTED) but
> not yet committed(t_xmin > CurrentTransactionId) tuples.
> He will see them in the future.
>
P.S.
This is the main reason that I once proposed to call
'pg_ctl stop' to stop postmaster in pg_upgrade before/after
moving pg_log and pg_varibale.
There was a dicussion to recycle OIDs.
It's impossible to recycle XIDs.
Regards.
Hiroshi Inoue
