traffic_manager dies when traffic_server dies

This seems to be some sort of regression / breakage, it does not reproduce with 5.3.x. What seems to happen is that if traffic_server crashes, traffic_cop always kills traffic_manager as well. I see e.g. this in the logs:

Aug  5 14:34:46 qa1 traffic_cop[13926]: (cli test) could not communicate with mgmt cli
Aug  5 14:34:46 qa1 traffic_cop[13926]: could not contact manager, assuming server is down