The context manager name refered to by the cluster doesn't appear to be reset when the context reloads. Other nodes in the cluster receive messages with an incorrect manager name, after a distributable context is updated by adding or updating a jar or the web.xml file. The name appears to be appended to it's previous value after each reload. Other members of the cluster log messages from the updated node as follows: SEVERE: Context manager doesn't exist:host1host1 SEVERE: Context manager doesn't exist:host1host1host1 (In our case there was only a root context in each host.) The Cluster definition is exactly as it is in the default server.xml, except that it's placed at the Engine level, rather than inside a Host.
I see what you mean, and I traced it to the getManagerName method in org.apache.catalina.cluster.tcp.SimpleTcpCluster. I think I've put in a simple and safe fix in SVN right now, but I'm not the clustering expert, so it'd be great if someone else could test it.