Steve

I will try the log shipping and let you know.

Thanks


On Mon, 2011-06-27 at 09:26 -0400, Steve Singer wrote:
> On 11-06-26 10:48 PM, rodney wrote:
> > Hi
> >
> > I'm trying to setup slony to replicate table from one master to around
> > 200 slaves. Can slony support slave number at this scale? It is observed
> > in pgadmin that when I have 1 slave, the num of listener is 1. However,
> > when I have 2 slaves, the num of listener increase from 1 to 4 (not 2).
> > Will this be a problem if the slave go to a large number, like 100?
> >
> >
> > Best regards
> 
> Running 200 slaves isn't a common use case for slony.
> That doesn't mean it won't work.  I know of at least one person who is 
> doing about 50 slaves, but I don't have details on the setup.
> 
> If your going to try Slony with 200 slaves this would be my advice
> 
> * Expect to encounter issues along the way, you will need to figure out 
> the best way to deploy and tune things
> 
> * Slony supports the idea of cascaded subscribers.   You can have node 
> something like
> 
>      1
>      |
>      V
>      2
>     / \
>    3   4
> 
> so you can have a number of direct subscribers then feeding a number of 
> slaves.   Creating a layered network would probably work better than 
> having 50 or 100 subscribers off a single node.
> 
> * Slony also supports something called 'log shipping' where the slon on 
> a slave generates .sql files that can then be applied to other nodes 
> downstream.  Since these other nodes are not technically part of the 
> slony cluster (they don't have paths, or slon daemons) you can 
> distribute the .sql files to as many machines as you want.  This might 
> work for you
> 
> * In a slony cluster each node must confirm SYNC messages from EVERY 
> other node even when a direct path between them doesn't exist (this 
> explains  for the number of listeners you see in pgadmin).  At a 100 
> nodes there will be a lot of SYNC/confirm messages.  You will probably 
> have to increase the SYNC interval so you have fewer syncs going on.
> 
> 
> * If you try any of this let us know what works or doesn't
> 
> Steve
> 

-- 
rodney <yl.wang at invantest.com>