Re: Problem with pg_upgrade (8.4 -> 9.0) due to ALTER DATABASE SET ROLE

On Jan6, 2011, at 04:13 , Bruce Momjian wrote:
> Robert Haas wrote:
>> On Wed, Jan 5, 2011 at 9:44 PM, Bruce Momjian <bruce@momjian.us> wrote:
>>> I think pg_dumpall would have failed with this setup too, so I don't see
>>> this as a pg_upgrade bug, nor something that I am willing to risk adding
>>> to pg_upgrade.
>> 
>> If adding RESET SESSION AUTHORIZATION fixes the bug, I think we should
>> consider doing that.
> 
> If we add every fix that could conceivably break a pg_dumpall restore,
> pg_upgrade will be less stable than it is now.  I don't see why adding
> this should be any different.

The issue is more complicted. In my situation, it's not the pg_dumpall
restore that's failing, but rather pg_upgrade's attempt to install
the support functions necessary for the upgrade.

But in principle, you're right I think. pg_dumpall *would* fail if my
database contained any objects that required superuser privileges to
create, like C-language functions. 

> If you want to argue that pg_dumpall should be doing it, that is a
> separate issue and not related to pg_upgrade.

I think both need the fix.

best regards,
Florian Pflug