"Matthew T. O'Connor" <matthew@zeut.net> writes:
>> [ pg_upgrade won't be able to change user table representation ]
> How limiting is the above? Does this mean that pg_upgrade will be
> rendered invalid if there is an on-disk representation change? Do we
> think we will make it from 7.4 -> 7.5 without on-disk changes? Do we
> think at this point most upgrades will be without on-disk changes?
Per prior discussion, we will enforce some sort of limit on how often
the representation of user tables/indexes can be changed. The idea will
be to "batch" such changes so that you only have to do a dump/reload
every N major releases instead of every one. In other words, pg_upgrade
will work for most version upgrades but we reserve the right to
occasionally make releases where it doesn't work.
How large N will be in practice remains to be seen, of course, but I'd
expect something on the order of 4 or 5.
In theory pg_upgrade could be made to apply changes in user data
representation, but I'm unconvinced that such a process would be a big
improvement over dump/reload.
regards, tom lane