On Fri, Nov 21, 2008 at 5:47 PM, Shreedhar Ganapathy
<Shreedhar.Ganapathy_at_sun.com> wrote:
>
>
>
> I can see multicast traffic ... related to cluster events (instances stopping/starting)
> but nothing that looks like a session being replicated (or copied
> over) - even when I
> tweak the load balancer to send a client to one node, have them log in
> and then point the LB to the other node.
> Can you restate the steps you followed for the above to help our folks
> reproduce this behavior?
> I assume you are following the High Availability guide.
Yes, this is mostly out an out-of-the-box GF cluster, with a Citrix Netscaler
(HTTP switch) in front of it.
> Just to reaffirm. Your app is availability-enabled and if its a web app it
> has the <distributable/> tag in the web.xml.
Yes to both.
> And at the moment you are using clusterjsp web app. Right?
No, this is a deploy of the Roller weblogger engine. I'm going to try
clusterjsp on the development
cluster tomorrow to see if that has the same issues, I think.
> Do you see any sessions replicated at all or none?
Nothing obvious in the logs, but then no errors either.
Hence me asking for debugging tools :)
> Session replication component uses GMS only for cluster lifecycle events
> such as an instance joining, leaving or failing.
>
> In-memory replication component also uses Jxta directly for its replication
> traffic using Jxta's JxtaBiDiPipe which is an bidirectional socket
> abstraction.
Ah, hadn't realised that; assumed session replication piggybacked on
GMS traffic.
Does that mean sessions do not go over multicast?
> Does anyone know of any introductory reference material on writing a
> GMS client?
> http://blogs.sun.com/shreedhar/entry/shoal_clustering_101
Thanks, that's very useful.