The main loop in tl-xinit waits for changes of subprocesss exits and signals (client and server). When client exits/signaled the main loop is exited and a cleanup of session is performed. However when server exits or is signaled we assume that the client will die due to X is killed which is not the case if client processes is not dependent on X server, such as upstart on ubuntu. lightdm and gdm does not assume this behaviour and reacts when X server is exit/signaled.
Here follows xinit.log for profile with only a xterm which will die when X server is gone. Executing start program command: xterm xterm: fatal IO error 104 (Connection reset by peer) or KillClient on X server ":1" XIO: fatal IO error 11 (Resource temporarily unavailable) on X server ":1" tl-while-x11: lost Xserver connection, terminating child 9100 ... after 116 requests (116 known processed) with 2 events remaining. tl-while-x11: lost Xserver connection, terminating child 9106 ... tl-xinit: warning: Xserver terminated and returned 0 Profile command exited with exit code 84 Running /opt/thinlinc/etc/xlogout.d/tl-kdestroy.sh Running /opt/thinlinc/etc/xlogout.d/tl-umount-localdrives tl-xinit: client terminated and returned 0 tl-xinit: Session terminated. Exiting. Here is a unity profile which never will die, tl-xinit still lives after kill av X server. Executing profile: unity Using XDG session: ubuntu Executing XDG session command: upstart --user XIO: fatal IO error 11 (Resource temporarily unavailable) on X server ":1" after 116 requests (116 known processed) with 0 events remaining. tl-while-x11: lost Xserver connection, terminating child 9403 ... tl-while-x11: lost Xserver connection, terminating child 9409 ... tl-xinit: warning: Xserver terminated and returned 0
When a system administrator wants to disallow running programs or services in disconnected user sessions, they would append "-MaxDisconnectionTime <time>" to /vsmagent/xserver_args, as described in our documentation (https://www.cendio.com/resources/docs/tag/configuration_limiting_lifetime.html). As we never ensure from tl-xinit that the xsession process actually terminates when Xvnc does, which means that the user processes can live on as long as they handle/don't care about missing a display. As a system administrator, I would have expected that the user session was terminated, not just Xvnc. I can't remember hearing about this use case in earlier discussions, so I am reassigning this bug from 'Future' to '---' for further discussions.
*** This bug has been marked as a duplicate of bug 4753 ***
