[jboss-user] [Clustering/JBoss] - Re: Cluster is loosing his master

[bstansberry@jboss.com]

The use of JNDI port is not meant for prettiness.  It's actually intended that in the use case you describe in your initial post, the name should *not* be unique.

The idea here is to detect the case you describe, where a node is killed and restarted and rejoins the group before the group knows the first version is dead.  The intent is to prevent that happening by giving the node the same name and taking appropriate action (IIRC throwing exception) if the node name is a duplicate in the view.

In JGroups 2.6, a feature called "Logical Addresses" will be added that will deal with this problem at the JGroups level, avoiding this admittedly hacky generateUniqueNodeName business.

Your report indicates that this mechanism is breaking down; that's the bug.

BTW, does your JGroups config in cluster-service.xml include FD_SOCK next to FD?  If not, adding it is a good idea; with it node_a would detect the kill -9 of node_b immediately.  See http://wiki.jboss.org/wiki/Wiki.jsp?page=JGroupsFD_SOCK.

View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4039390#4039390

Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4039390