Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> Excerpts from Bruce Momjian's message of mié ago 08 17:15:38 -0400 2012:
>> On Wed, Aug 8, 2012 at 04:23:04PM -0400, Robert Haas wrote:
>>> I think this is one good idea:
>>> http://archives.postgresql.org/message-id/29806.1340655654@sss.pgh.pa.us
>> If we currently require 14 steps to use pg_upgrade, how would that
>> reduce this number? What failures does it fix?
> The suggestion by Tom reduces the list by two steps because it doesn't
> need to adjust pg_hba.conf or put it back in the original way
> afterwards.
Even more to the point, it flat-out eliminates failure modes associated
with somebody connecting to either the old or the new cluster while
pg_upgrade is working. Schemes that involve temporarily hacking
pg_hba.conf can't provide any iron-clad guarantee, because if pg_upgrade
can connect to a postmaster, so can somebody else.
The point I think Robert was trying to make is that we need to cut down
not only the complexity of running pg_upgrade, but the number of failure
modes. At least that's how I'd define improvement here.
regards, tom lane