On Dec 9, 2007, at 6:09 PM, Peter Geoghegan wrote:

>
> 1. Make changes to the database with no Slon daemons running,  
> through my C++ front end.

Change this to leaving the slon on the origin server running, or at  
least use the script that generates sync events on the master while  
slon is down.

but your basic outline would work.  the trick is figuroing out how to  
start and when to stop the slons for each replica.  at least you can  
run them all on the same host -- no need to remote execute jobs.

One possibility would be to check the sl_status data for your remote  
node.  Note the st_last_event value, and monitor for when  
st_last_received exceeds that number. Then you know you're as up to  
date as of the time you started the slon, and can kill it off.  That  
assumes all the updates you're interested in have been done by the  
time you start slon for that node.