We are running multiple Open Ondemand servers on the backend to HAProxy 1.8. Interactive desktops running on cluster nodes are freezing up at different times even with activity on the websockify socket to the end client. I have a heartbeat setup on the websockify socket and a ping to another cluster node running and at different times the desktop freezes up. I can exit the session and reconnect to the websockify socket and the session is still running on the compute server. I currently have the ‘timeout tunnel’ set in defaults to 120m and get these freeze ups many times in a 2 hour session. In a 1 hour session it will freeze up at about 20 to 30 minutes into the session. The logs only show the original connection to the websocket and nothing after that until I reconnect to the socket after the freeze up. Any ideas what I can set to keep these websockify websockets from freezing up?
This was due to the fact that the per user file descriptors was set to 1024. Adding a ulimit-n 65536 to the conf file has fixed this.