Re: Vaccuming dead rows on busy databases

Tom Lane wrote:
> "Greg Sabino Mullane" <greg@turnstep.com> writes:
>> I'd like to see something like VACUUM FULL WAIT. :)
> 
> Sounds like a deadlock waiting to happen :-(
> 
> AFAIK the general practice is to just accept the fact that vacuum can't
> remove recently-dead tuples.  You should look into whether you can't
> shorten your transactions --- very-long-running transactions create
> other performance issues besides vacuum not removing stuff.

It seems to me that the most common support problem I keep seeing on the 
mailing lists is VACUUM not working well because of long running 
transactions.  If I understand it correctly, people have talked about 
reducing the problem by tracking xmin (or something, sorry if I'm 
getting this wrong) on a per table basis rather and per cluster.  Now 
I'm sure this is not simple and I know I don't have the skills to do it, 
but I think it would resolve (or at least significantly mitigate) what I 
perceive as one of the biggest usage problems with PostgreSQL.

Comments?