Not all Fluent session exit on Python exit if watchdog is not used

[mkundu1]

Script to reproduce:

import ansys.fluent.core as pyfluent
solver1 = pyfluent.launch_fluent(show_gui=True, start_watchdog=False)
solver1.scheme_eval.exec(['(pp "solver1")'])
solver2 = pyfluent.launch_fluent(show_gui=True, start_watchdog=False)
solver2.scheme_eval.exec(['(pp "solver2")'])
solver3 = pyfluent.launch_fluent(show_gui=True, start_watchdog=False)
solver3.scheme_eval.exec(['(pp "solver3")'])

In my trial, the last session solver3 keeps on running. Found while investigating an asynchronous launch_fluent behaviour for #1726. I think this is an important behaviour to maintain for the correctness of session launch code.

[mkundu1]

The issue disappears if I disable starting streaming services during launch_fluent.

[mkundu1]

This is also reproducible with a single session.

The issue is related to #1358's changes. session.exit() is handled in a daemon thread (MonitorThread) which ensures shutdown of all streaming (non-daemon) threads. A daemon thread is terminated abruptly during interpreter exit. Earlier the streaming threads were shut down after the server was exited within session.exit(), so the server-exit was guaranteed by MonitorThread. But after #1358, this sequence is reversed, the MonitorThread shuts down immediately after shutting down the streaming threads without exiting the last server.

#1358 fixes the issue that on initialization event from server, a new monitor streaming call is initiated which causes a hang if the server is already exited.

We need to find a solution that satisfies both requirements.