Bruce Momjian <bruce@momjian.us> writes:
> On Sun, Sep 16, 2012 at 12:38:37PM +0800, Rural Hunter wrote:
>> I ran the pg_upgrade with the patch and found the problematic object
>> is a toast object.
> OK, this is exactly what I wanted to see, and it explains why pg_dump
> didn't show it. Can you find out what table references this toast
> table? Try this query on the old cluster:
> select oid, * from pg_class WHERE reltoastrelid = 16439148;
> I believe it will have an oid of 16439145, or it might not exist.
Most likely what's happened is that the table has a toast table that
it doesn't need, as a result of having dropped the only wide column(s)
in it. So when the table is recreated in the new cluster, there's no
toast table for it.
So what you need to do is get rid of that check, or relax it so that it
doesn't insist on toast tables matching up exactly. It seems possible
that there could be discrepancies in the other direction too, ie,
new cluster created a toast table when old cluster didn't have one.
regards, tom lane