Andrew Sullivan <ajs@crankycanuck.ca> writes:
> On Thu, Nov 04, 2004 at 10:00:23AM -0500, Tom Lane wrote:
>> If you read the code a little more closely, you'd see that it already does.
> Hmm, so obviously I was confused in my other message. But I've seen
> the same sort of effect as the OP: transactions in another database
> on the same back end seem to prevent some recovery by vacuum in the
> local back end. Is this just an illusion?
I think it's most likely that there were also old transactions in the
current database. Only the shared tables (pg_shadow, pg_database,
pg_group) are vacuumed using a cutoff that depends on non-local
transactions.
Looking at the back versions, it appears this logic was put in in 7.2;
is it possible you are remembering the behavior of older versions?
regards, tom lane