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

Bruce Momjian <bruce@momjian.us> writes:
> We could modify pg_dump to emit RESET AUTHORIZATION in --binary-upgrade
> mode.  I am unclear if that might cause some other problems though.

I finally figured out what was really bugging me about that proposal:
it's a one-shot hack for fixing one problem that could arise from
non-default ALTER DATABASE/ALTER ROLE settings.  Who's to say there
are not other such issues, either now or in the future?

It occurs to me that a more principled way to deal with this class of
problems would be to delay restoring ALTER DATABASE/ALTER ROLE
settings until after everything else is done.  Not sure what the
implementation consequences of that would be.  Ideally we'd make
pg_dumpall output work that way in general, not just for pg_upgrade.
        regards, tom lane