On Mon, Jan 03, 2005 at 07:54:15PM -0500, cbbrowne at ca.afilias.info wrote:
> > I'm testing a 3 node Slony-I cluster, and it replicates without problems,
> > until I try to do a switchover as described in slony-I-failover.txt, by
> > using the following slonik commands:
> >
> > lock set (id = 1, origin = 2);
> > wait for event (origin = 2, confirmed = 1);
> > move set (id = 1, old origin = 2, new origin = 1);
> > wait for event (origin = 2, confirmed = 1);

> OK, so can I make a few assumptions and see if they're right?
> 
> --> I assume that you didn't see any manifest errors in the process of
> moving the set origin.
> 
> --> I assume that you're seeing activity go thru fine between nodes 1 and
> 2, and you're simply not seeing _anything_ happen on node 3.  It sorta
> looks like node 3 is OK; it just isn't getting updated, right?
> 
> If those are all good assumptions, then the problem is almost certainly
> that you have a broken/inadequate set of "listen paths."  They're in table
> sl_listen, set up by the STORE LISTEN command.
> 
> This is listed in the FAQ; you can see a description of a similar scenario
> here:
> <http://linuxdatabases.info/info/faq.html#AEN41038>

I ended up creating all possible listen path's (receiver != provider), and now
it works.

But when moving origin of the set from node 1 to node 2, I end up with
having node 1 forward all changes to node 3. And what I really need is a way
to take down node 1 for maintenance, and that I cannot do while it forwards
to node 3.
I can only see 2 ways to change that - either I can unsubscribe node 3 and
subscribe it with node 2 as provider, but that locks down node 3 for quite a
while, or I could do a failover instead, and risk loosing some data.

--
David Darville