[pulseaudio-discuss] ubuntu usb "desktop-server" with single user and constant audio feed from mpd

[Martin Hamant]

Le 16/09/2011 13:35, Colin Guthrie a écrit :
> 'Twas brillig, and Martin Hamant at 16/09/11 10:19 did gyre and gimble:
>> Hi there,
>>
>> I need to setup a server that permanently plays audio by mpd thru
>> pulseaudio. With or without X/gnome started. Means pulseaudio and mpd
>> should start before the graphical part.
>> There will be only one unique user on this box.
>>
>> using Ubuntu 10.04 LTS, pulseaudio daemon starts thanks to the
>> auto-spawn feature, when a gnome user logs in -  the issue is,  if I
>> start mpd (or any software that use pulseaudio) before gnome, it'll
>> spawn a pulseaudio daemon that belongs to the mpd user. So I get two
>> pulseaudio instances: one which belongs to mpd, the other to my unique
>> system user... But here I need my system user to "supervise" volumes and
>> pulseaudio properties for the whole system !
>>
>> if I delay mpd startup (by modifying init.d script....), all's going
>> well - as it connects to the existing pulseaudio instance. But I don't
>> feel good about that  "strap".
>>
>> In a common usage you won't really notice what I describe here, but as I
>> am running this ubuntu from a USB stick, there is a period of time
>> before gnome starts where any program that use pulseaudio will start its
>> own instance.
>>
>> Of course the solution that comes in mind is System Mode. Not good.
>> Second: delays =>  not good
>> So... What is the best pratice to do that ? Is running mpd as the user
>> which runs the gnome session is a working /or good solution ?
>>
>> Thanks !!
> OK, so the long term solution is for Ubuntu to switch to systemd (I'm
> sure they will eventually) and then use the login lingering feature of
> session management to allow certain user processes to start before the
> user has logged in (this is specifically designed for situations like
> sharing files via network with Rygel).
>
> This feature could be extended to PA but it does come with some
> complications. Only the "active" user is allowed access ot audio. When
> the login manager is present this is GDM (of course Ubunutu are also
> potentially going to use lightdm which will be broken in all sorts of
> ways until they start running a full session (which I think was the
> whole reason to use it in the first place... so don't know what's going
> on there).
>
> Now if the login manager wants to use sounds (e.g. for feedback/earcandy
> etc.) then it needs to have it's own PA.
>
> But, regardless of the above complications, running mpd as a user
> lingered processes is likely the best route forward.
>
>
>
> As a secondary solution for you now, what I'd recommend is the following:
>
>   1. Add your user to the "audio" group. THis means you'll always have
> the right to use the audio devices on your system even when not logged
> in. This comes with some drawbacks (like gdm user not being able to
> access the device when you've got it open with mpd) but in a single
> (real) user system, this isn't too bad.
>
>   2. Run mpd as you, not as an "mpd" user.
>
>
> With this setup things should (mostly) work OK for you.
>
>
> The alternative is to use system wide mode but then you won't get the
> benefits of e.g. SHM and module loading etc. etc. without doing quite a
> few tweaks.
Thank you Colin for answering that fast.
"run mpd as you" implies to not run mpd as a system wide service but 
with "manual" start.
Or I would have to modify system wide /etc/mpd.conf with my user: that 
would work but would be a little inconsistent...

In another hand, the user isn't doing anything else that playing audio 
files thru mpd, and recording with another app: liquidsoap which runs as 
'liquidsoap' user, which is in the pulse-access group..... system sounds 
are disabled.
The question is: do I need on the fly module loading in that context ?

Thanks again !!