Alvaro Herrera <alvherre@alvh.no-ip.org> writes:
> There has been talk in the past about creating a tool for in-place
> upgrades. The most complete description of the needed work I've seen
> posted to date is here:
> http://archives.postgresql.org/pgsql-hackers/2003-12/msg00379.php
> There was another proposal after that,
> http://archives.postgresql.org/pgsql-hackers/2004-09/msg00916.php
> Neither of them yielded any useful software, that I know of.
Yeah, I'm afraid I got distracted and never got very far on the first
scheme. It still seems perfectly workable, within the limited goals it
has --- in particular, it doesn't support changes in the on-disk format
of user tables; hence no tuple-header-layout changes or datatype
redesigns. But it would work for 99% of the changes we make, and
certainly for all cases that seem plausible to happen during beta
cycles.
I think the second one isn't worth the electrons it's written on,
because the smgr API is at far too low a level of abstraction to be
able to solve the problem :-(
regards, tom lane