On Mon, 2008-07-14 at 21:59 +0200, Kaare Rasmussen wrote:
> > I just lost a months worth of stats data myself, so join the club. It
> > wasn't critical data, but it would have been nice to have kept
> > around...
>
> I also think there could be a TODO item in it. If vacuum instead of removing
> items, somehow stashed them away in a storage limited archive it would be
> possible to do a SELECT...AS OF TIMESTAMP.
>
> The idea is of course to be able to retrieve rows that really are deleted, but
> are still on disk as non-vacuumed or vacuumed and not removed completely. And
> it would also take a 2. stage vacuumer to keep the storage within its limits.
I've got the design all worked out for this.
The "only" thing we need is a VACUUM that will remove unseen data from
within the middle of the sum-of-all-snapshots, if there is a gap. At the
moment we never remove rows beyond global xmin, but we could iff the
transactions at xmin promise never to update data. That should go on the
TODO list as a precursor. Some discussion required :-)
-- Simon Riggs www.2ndQuadrant.comPostgreSQL Training, Services and Support