On Wed, Jun 17, 2009 at 05:17:13PM +0100, Ben wrote:

> I recently tried adding a second slave node (3) to this cluster, as a  
> cascaded slave from node 2. While the replication *worked*, we were  
> seeing some unexpected behaviour on the other two nodes - I wanted to  
> ask whether this is usual, and just a misunderstanding on our part as to  
> how cascaded replication works.

Nope, can't do what you want.

>
> The intention was to set up node 3 with no direct access to the master  
> node (1)

This isn't really the way it works.  Slony is designed so that node3
could get its data from node1 if node2 failed permanently -- then
you're not totally hosed.  If you want node3 really never to be able
to talk to node1, then you need to do log shipping from node2-> node3.
This is not, I realise, a perfect solution, and it doesn't cover a
possible use-case that Slony might otherwise cover (sort of a bitter
irony, considering how complicated Slony is in order to cover every
other use-case), but it's the way things are.

There are some very tricky things you can do with the paths such that
you might be able to hack around this, but it's not a natural feature
of Slony.  It might be interesting to design a feature to make this
more natural, though.

A

-- 
Andrew Sullivan
ajs at crankycanuck.ca