I've got a customer [1] that started receiving the following error during multicast sends (from MERGE in this particular log) until JBoss was restarted:
[Email Dennis]
[org.jgroups.protocols.UDP] [ERROR] failed sending message to null (0 bytes)
java.net.NoRouteToHostException: No route to host: Datagram send failed

I've had one other case with this symptom. In that case, they had bonded network interfaces and were testing failover by bringing one side of it down.

It looks to me like something is screwed up at the OS level that doesn't get fixed until JBoss is restarted (or the socket is recreated?).
Is there anything we could do about this at the JGroups level? The customer has asked why we don't recreate the socket when this exception happens. But that doesn't seem like a good solution to me since there's nothing about a NoRouteToHostException that should indicate the socket needs recreated.

-Dennis

[End email Dennis]

One thing we could investigate is to close and recreate a socket when such an exception is encountered. Possibly other exceptions as well...
This would affect the transports (UDP and TCP), plus some of the protocols which use sockets, e.g. MPING, FD_SOCK, STREAMING_STATE_TRANSFER etc...