Re: Vacuum and oldest xmin (again)

On Thu, Nov 04, 2004 at 09:31:05AM +0100, Kuba Ouhrabka wrote:
> initial data loading are essential tasks. The only solution I can see
> now, is to have several database clusters on the server in order to have
> completly separated databases...

We actually do that, for the reasons you say, plus because it gives
us a certain degree of separability (and because it allows us to tune
the caches more effectively for each type of system).

> My suggestion is to add some more logic to vacuum to get correct oldest
> xmin - local to current database. 

I think the problem is that the xids are in fact global values.  This
is, importantly, why you get messages about not having vacuumed in a
long time in case you have a database which is not in your regular
vacuum regimen.  I have my doubts that the idea of the xids "local to
current database" is even a coherent idea in Postgres, but I may be
wrong (in which case someone is bound to correct me).

A

-- 
Andrew Sullivan  | ajs@crankycanuck.ca
When my information changes, I alter my conclusions.  What do you do sir?    --attr. John Maynard Keynes