1. Don't listen to more than one port through the cluster. Instead, listen on the lonePort as a master, not as a worker, so that it doesn't share more than one fd with the master. 2. Handle race condition in process.disconnect() in the worker, vs worker.disconnect() in the master. 3. Remove the close listener for the other server when shutting down, rather than trying to close everything every time.
They're invalid anyway
Wtf? Not sure what's causing this, but saw it in the logs.