Brad Nicholson wrote:

>Marc G. Fournier wrote:
>
>  
>
>You use 1 and 2.  The provider/reciever are for the sets, not the 
>nodes.  The procedure that you'd want to follow is
>
>1)Create set 999 on node1
>2)Subscribe set 999 to node 2 (provider is node1, receiever is node 2)
>3)Merge set 999 into set 1.  This will put the table from set 999 into 
>set 1, and set 999 will be gone.
>
>If you were to use node 3 and 4, I beleive the merge set would fail.  If 
>memory serves me, the two sets have to have the same providers and 
>recievers across all nodes before you can merge them.
>  
>
It would actually break long before that.

Suppose we have cluster MINE with hosts h1 and h2, with databases on
port 5432, and each uses database my_slony_db...

Creating node 1 establishes namespace _MINE on the database on h1.
Creating node 2 establishes namespace _MINE on the database on h2.

TRYING to create node 3 on h1 would TRY to create namespace _MINE again,
which would fail, because it's already there.  Ergo, there can be no
node #3.  Ditto for trying to have node4 on h2.

So no, you don't create extra nodes for this.

Remember, everyone, a node is a database participating in replication. 
Adding in additional sets does nothing to change that essential fact...