SocketsAndInterfacesTestCase.getNonDefaultNic() is unreliable as it scans the system interfaces and picks the first one that meets minimal criteria. It's not at all certain that the one it picks is suitable for binding an undertow listener.

If -Dipv6 is used the test doesn't do this and just uses the node0 nic. That bit was added when the test was 5 years old. That seems to work fine and I can't find anything that justifies not always doing that. There's nothing about the test assertions that explain why a 'non-default nic' is important.