Clustered SSO

[marcreis]

I have run some tests for performance measurements and have noticed that in a cluster of two 4.0.4GA (Apache 2.0, mod_jk 2.0.55, Sticky Sessions) when simulating a failure (shutting down the server, to which the connection has been assigned) and then failing over to the other node of the cluster has mad login on this node necessary. Sometimes this seems to work and sometimes it doesn?t. I haven?t bee able to spot what could be the cause. I would have expected that this is never needed with the SSO active. Is there a situation that could enforce this? Like heavy load not allowing the SO to be ?transmitted? to the other node and so failing when failing over?

Sincerely

Marc

(The config is that of the original server.xml and jboss-service.xml with the mod_jk changes and just Clusterd SSO uncommented)

[brian.stansberry]

Perhaps you are seeing http://jira.jboss.com/jira/browse/JBAS-3096?

[brian.stansberry]

Also, the JBoss Cache instance used for HttpSession replication and ClusteredSSO is configured for asynchronous replication. This means it is possible for an SSO to be started on one server and then that server dies before the SSO is replicated. The heavier the load, the greater the possibility, although still not high.

[marcreis]

Hi Brian,
thanks for the reply !
I checked the JIRA you linked. Not sure which of the two problems is showing up, but I guess that it is one of them and knowing that all is ok.
I belive that it could be due to the load, since I am stressing the AS somewhat and it is running in a XEN with two virtuell pc's. Since there are NAKACK messages showing up, the load seems to be very likely for causing this.

2006-09-25 18:31:43,490 WARN [org.jgroups.protocols.pbcast.NAKACK] [06-2:34224] discarded message from non-member 06-1:33572

Thanks !
Sincerely

Marc