Re: vacuum vs open transactions

On Wed, 2005-01-12 at 11:59, Ed L. wrote:
> I'm looking at some 7.3.4 vacuum output, and at first glance it does not
> appear that vacuum is reclaiming any dead tuple space if there is even a
> single open transaction, even if the open transaction does not in any way
> reference the table being vacuumed.  Is that correct?  Is the behavior
> different in later versions?

I believe the problem is occurring if the open transaction is older than
the tuples that could be vacuumed.  The MVCC system means that as long
as a transaction that started X hours ago is still open, the tuples that
have been freed since then can't vacuumed because they need to stay
visible for that transaction.