On Wed, Aug 19, 2009 at 11:12:26PM -0700, roctaiwan wrote:

> My node 1 (master) has all the sets from 1-5. node 2 has subscribe to set 1
> and 2, node 3 subscribe 3, 4, 
 
> If I do switchover from node 1 to node 2, if I am correct, node 1 will
> forward all the sets to node 2, node 2 then will SYNC with all other nodes
> (or all from Master) to collect all datas that it does not have, preparing
> itself to become same "status" as the original Master(node1) and now node 1
> the original master has become the subscriber to node 2. 

No, I don't believe this is right, assuming I understand you.
Switchover works on sets, so you will effectively make node 2 the
origin for sets 1 and 2, but sets 3-5 will remain origined on node 1.

The key thing here to get is that Slony has no real notion of "all
nodes" involved in replication from one node to another.  It's just
not interested in that.  At the basic level, it is set-oriented.  So
the sets are nodes involved in replicating that set.  You can (could?)
conveniently work on supersets using external tools, but Slony doesn't
itself care about that.

You raise an interesting point about the documentation, however, which
is that it doesn't make clear that the new origin (in the case of MOVE
SET) has to be actually subscribed before moving to it.  This is
similarly unclear in the FAILOVER command, although I suppose in that
case it ought to be obvious that if you're not already subscribed,
failing over to such a node is doomed.

A
-- 
Andrew Sullivan
ajs at crankycanuck.ca