On Thu, 2006-10-19 at 10:21 -0400, Andrew Sullivan wrote:
> On Thu, Oct 19, 2006 at 09:11:13AM +0200, Barry wrote:
> > Hi
> > 
> > I have been running this for 3 days now,
> > and still the db has not caught up on the old pre-replicating data.
> > 
> > Is there somewhere I can look to confirm that this will take place ?
> > Should it take this long ?
> 
> Hrm.  Which version is this?  There was a problem at one point where
> you had to pay the retail cost of indexes when restoring, so you had
> to manually drop all the indexes on the replica prior to building it,
> or it'd basically never catch up.  I don't recall just now when that
> was added as an automatic feature.  Check the docs for your release.
> 
> You can check that the COPYs are happening on the replica, though,
> easily enough, with ps.  
> 
> Three days does seem rather a long time.

My suspicion is that something is interrupting the copy set.

Check the slon logs to see whats going on (you want the slon log for the
node that is provider for the set).  You'll see messages that relate to
the start of the copy set, then a messages indicating how long each
table took to copy.  Once it completes, you'll see the total time for
the copy set.

If you see the message about starting the copy set multiple times,
that's where you need to look for trouble.  Something is interrupting
the copy set (likely the slon falling over).

Oh, if this is the case, you target node will have a ton of dead tuples,
you'll need to vacuum it (possible a vacuum full if your FSM is clobber,
which it may be).

What I'd do if that's the case, - vacuum the target db, kill any of the
slon watchdog process if they are running, and try re-subbing.

-- 
Brad Nicholson  416-673-4106
Database Administrator, Afilias Canada Corp.