On Mon, Oct 23, 2006 at 03:51:30PM -0600, paul cannon wrote:
> Our installation of slony 1.2.0 is struggling.
> 
> It started trying to do a log switchover several days ago. Since then,
> every 10-20 minutes we get this in the logs:
> 
> NOTICE:  Slony-I: log switch to sl_log_2 still in progress - sl_log_1 not truncated
> 
> Between each time, it looks like it's doing a DELETE on both logs and a
> VACUUM on sl_log_1.
> 
> sl_log_status on the master node is 3, which from the code means
> everything should be going into sl_log_2, right? And yet /both/ logs are
> continually growing, aside from the periodic DELETEs.

I realized that everything going in to sl_log_1 was characteristic of a
few certain applications, and that those applications had connections to
the db dating from before the slony upgrade to 1.2.0.

When I restarted those applications to make them reconnect to
postgresql, everybody started behaving and inserting into sl_log_2 only.
The log switchover finally completed successfully, and sl_log_status is
now 1, as expected.

So I guess the slony stored procedures were being cached or something,
or updates do not become visible in parallel connections? That's an area
of postgres about which I know very little.

All in all, it may be a good idea for people upgrading to slony to
reconnect any applications which submit data.

-- 
paul