On 25.04.2012 18:49, Robert Haas wrote:
> Maybe this is a silly idea, but if you're thinking about creating a
> local XID space and a global XID space, it might be a good idea to
> also make allowance for an "unlogged" XID space - that is, an XID
> space that is global to all backends but need not survive crashes.
> This would potentially allow unlogged tables to be used in HS mode. I
> would expect that you probably don't want to go as far as actually
> trying to make this work as part of your current project, but maybe as
> you're structuring the code it would be worth considering the
> possibility that we'll eventually want>2 XID spaces, either for this
> or other reasons.
It would also be nice to have a separate XID space for each database in
the cluster (plus one for global tables). Then one database consuming a
lot of XIDs would not force anti-xid wraparound in other databases, and
it would take us one step closer to being able to (physically)
backup/restore/replicate a single database in a cluster.
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com