On Tue, 2007-12-18 at 00:48 +0100, Lennart Poettering wrote:
On Sun, 16.12.07 09:46, Richi Plana (myfedora(a)richip.dhs.org) wrote:
> Again, ideally, services (including system ones) should 1) only be
> started and 2) actually be started when it is called for. Something
> similar to the way kernel modules are autoloaded when there's a need for
> the service. Or xinetd does for some services which use it (though with
> an option to actually stay resident).
Besides (very few) exceptions, kernel drivers are nowadays loaded when
the hardware they driver becomes available -- and not only when they
are used.
You can configure PA so that it is autospawned when needed. However, I
do not recommend this. I recommend to start it from the session
manager like we do it right now for KDE and GNOME. Why? Because PA
nowadays does much more than just proxying access to the hw. It reacts
on hotplug events, network configuration changes, certain X11 events,
it is a network server, and so on and so on. For all these reasons it
is better to leave PA running all the time.
But how are those hotplug events communicated? That's what I meant when
I said the pulseaudio sound server should be started on an as-needed
triggering event. If there are too many entry paths to using PA, then I
understand the complexity.
What's the underlying communication mechanism for event passing? D-Bus?
TCP communication? For TCP sessions, we have xinetd. Would there be an
interest in having a similar service for spawning object / services as a
result of D-Bus service requests? Or is there already an existing
infrastructure? Now I'm just curious.
--
Richi Plana