Thanks for your replies, Andrew and Alan,

On Wed, Aug 20, 2008 at 10:04:09AM -0400, Andrew Sullivan wrote:
> 
> Wju are you running manual vacuums and autovacuum too?  You shouldn't
> need the full vacuum.  Anyway, assuming this is a release after 8.1 (I

Hmm, I'm not sure but I guess it could be a leftover from a previous 
PostgreSQL installation (ie. a pre-autovacuum release). Or maybe it's 
just because, as you said, autovac is not that reliable in 8.1 (yes, 
I use 8.1.10).

> > Thank you for the tip, it does rings a bell (ie. since I had an "almost
> > disk full" situation just before the replication problem, maybe pg may
> > have launched an emergency autovacuum of some sort? I need to explore).
> 
> No, PG won't do that (the emergency autovac happens in 8.3 if you're
> about to roll over your xid space); but it does suggest the table
> bloat I'm supposing, from inadequate vacuuming.
> 
> To solve it, you could do VACUUM FULL and REINDEX.  Just be prepared
> to wait a long while. 

I ran a "VACUUM FULL VERBOSE ANALYSE", and you were right about the bloat
(I had not complaints about fsm though). But this did not solve the
problem : the replication delta continue to increase (this node is now
over 5 days behind master). Well, I didn't dare to run a REINDEX, maybe
I should have. Or could this be a PostgreSQL or Slony-I bug ?


Thanks again.