That sounds very much like pg_listener bloat.

I guess you where not running pg_autovacuum, nor did any other regular 
vacuum maintenance on the two database. The latest versions in CVS, HEAD 
as well as REL_1_0_STABLE, have pg_listener added to the tables 
frequently vacuumed by slon.

Please issue a "vacuum full pg_listener" on both nodes and see if that 
gets things back to normal.


Jan


On 9/28/2004 11:36 PM, Justin Clift wrote:
> Jan Wieck wrote:
> <snip
>> What do you mean with "small update frequency"? Is it the interval in 
>> which the application does changes to the database or the interval slon 
>> issues SYNC events. I have not played with 100ms SYNC intervals yet, but 
>> I can imagine that it could create some sl_event and pg_listener bloat.
> 
> Well, I started slon with -s 100 -t 1200 (both nodes) in a simple master 
> + 1 slave setup and after about 2 days the replication "stopped".
> 
> Both slon's were still running, but no replication was occuring, and 
> looking at sl_confirm on the slave node showed no new rows being added 
> there.
> 
> Shutting down slon, then starting it up again (about 5 hours after the 
> timestamp of the last rows in sl_confirm) seemed to help, as a bunch of 
> new rows appeared in the sl_confirm table on the slave, but still no 
> actual replication was happening.  As evidenced by changing data on the 
> master and waiting for it to appear on the slave.
> 
> Shutting down the slon daemon on both master and slave, then starting it 
> up again (a few more times with no -s and -t values given on the command 
> line) helped and got things working again.
> 
> It's now running fine for the past day with an -s of 1000.
> 
> Not sure what caused the replication to stop, it just did.
> 
> No idea of the process for debugging it either.  All pointers would be 
> helpful of course.
> 
> :)
> 
> Regards and best wishes,
> 
> Justin Clift
> 
>> Jan
> 


-- 
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck at Yahoo.com #