[systemd-devel] systemd footprint

[Kay Sievers]

On Sun, Dec 4, 2011 at 20:45, Holger Winkelmann <hw at travelping.com> wrote:
> On Dec 4, 2011, at 6:15 PM, Kay Sievers wrote:
>> On Sun, Dec 4, 2011 at 17:52, Holger Winkelmann [TP] <hw at travelping.com> wrote:
>>> On Dec 4, 2011, at 4:40 PM, Kay Sievers wrote:
>>>>
>>>> For very simple setups, the D-Bus bus daemon is not absolutely
>>>> necessary, and can probably be made optional with a few changes, but
>>>> the D-Bus protocol is used by systemctl to talk to systemd, and can
>>>> not really be optimized out.
>>>
>>> systemctl talks directly to systemd via DBUS messages, or ONLY via dbus-deamon?
>>
>> Privileged users can talk directly over a private socket from
>> systemctl to systemd, unprivileged users need to go over the bus.a lot.
>>
>>> We have not much problems with DBUS concept, DBUS message formats etc. But may
>>> we can scale it down to a minimum for embedded use. May transport DBUS messages
>>> over regular sockets, TIPC, NETLINK sockets or something like brokerless ZMQ. something does
>>> not require another deamon and becomes network transparent
>>
>> There should be no need for that, D-Bus has private sockets, and
>> systemd uses that already. Systemd, as one can easily imagine, needs
>> to be able to start up and use its tools without relying on the
>> running D-Bus server.
>>
>>>> D-Bus is in not desktop centric, it's a plain IPC mechanism which lets
>>>> processes talk to each other. It's just that the desktop is not as
>>>> simple in its task as the usual base OS, so that it needs more
>>>> advanced technology like IPC and an object model to let components
>>>> integrate with each other
>>>
>>> Exatcly, thats what I mean, may we can remove some of the Desktop centric requirements
>>> and "just" use whats required in Server or embedded use cases. but keeping the vibrant
>>> ecosystem around DBUS and DBUS APIs.
>>
>> I don't see anything desktop centric in systemd. I have no idea what
>> you want to remove or configure-away here?
>>
>>>> For server OSs, and any other commonly used setup, there will be no
>>>> installation of systemd without D-Bus. It just can't work.
>>>>
>>>> Simple and limited embedded-like setups could work without the D-Bus
>>>> server, but still not really without the D-Bus protocol.
>>>>
>>> I agree. so lets call it eBUS a smaller DBUS use case,
>>
>> I don't think that makes much sense to have a name for that. You can
>> probably just change systemd to work fine without the D-Bus server,
>> but that would in no way be eBus or anything like that, it would just
>> simple be no D-Bus besides the wire format.
>
> Kay, Thanks for the clarifications about private sockets and such. Makes all
> more sense to me now. your are right, seems not much needs to configured out.

Sounds good. In case it wasn't clear, we are always interested in
clean patches that make such minimal stuff work. We usually don't
directly work with such very limited environments, but we surely
support the general idea and systemd should work on such small
systems.

Kay