> I think a consistent view on the slave during active replication is not
> quite guaranteed with this approach. Sequence updates are not
> transactional, we really don't know how to order them with respect to tuple
> updates. So someone reading the slave DB might possibly not see sequence
> changes appear in the order in which they occurred on the master. For our
> warm spare/slave needs, it appears adequate.
Aside from problems dealing with the loss of the communications link,
wouldn't it be better to implement a function call from the slave(s) to
the master to query the master's sequence? This can be done with pgperlu.
If there are INDEPENDENT sequences on the master and the slave, what's to
guarantee uniqueness?
--
Mike Nolan
