Back in 2013, in response to https://bugzilla.redhat.com/show_bug.cgi?id=963935 , behaviour was added to anaconda such that when a hub with no spokes is 'complete', the installer will proceed without any further interaction:

    def _update_spokes(self):
        from pyanaconda.ui.communication import hubQ
        import queue

        q = hubQ.q

        if not self._spokes and self.window.get_may_continue():
            # no spokes, move on
            log.debug("no spokes available on %s, continuing automatically", self)
            gtk_call_once(self.window.emit, "continue-clicked")

This was intended for initial-setup; the idea was to make initial-setup automatically quit if it wound up being run with no spokes (because user and root passwords were already set). Now, however, five years later, that's come back to bite us a bit, because of this feature:

https://fedoraproject.org/wiki/Changes/ReduceInitialSetupRedundancy

as part of that Change, the Workstation live installer no longer has any spokes on its install progress hub - both the user creation and root password spokes are suppressed. So even though install takes a long time, when it *does* complete, anaconda just auto-continues - which means it auto-quits, as this is the last hub. Basically, as soon as install completes, anaconda goes away and you're just left at a desktop.

Martin Kolman, Michael Catanzaro and I all agree that this probably isn't the best behaviour; users are likely to be confused and may assume that something went wrong, the installer crashed, etc. We believe that in this particular case - however it's achieved - anaconda should behave just as it does when there *are* spokes; it should print the "Installation complete" message and wait for the user to click Quit before actually quitting.

I'm proposing this as a Beta freeze exception, partly because it's a user-visible installer issue we can't fix with an update, but to be honest, *mainly* because making openQA work with this behaviour is actually quite difficult - if I can't figure out a decent way to do that, all openQA Workstation live install tests will fail, and the post-install tests that run from disk images created by the install test won't run.