Only allow one keepalive_timeout at a time. This prevents us from starting a second one and therefore losing track of the first when the server isn't responding within the 120 second timeout interval. Refs #4814; really shouldn't fix it, but it might.