Major Description
Under certain conditions, traffic_server does not seem to proxy the cop health check properly to the traffic_manager. Due to poor error handling in the manager thread, this can cause it to run indefinitely in a read() loop.
I think we should strengthen the code in manager, to check FD with a poll() before every read and write, and allow it to fail after a certain timeout (based on Cop's behavior) such that at least we end up killing the request and manager thread.