[OE-core] Conflict of systemd and sysvinit
Adrian Bunk
bunk at stusta.de
Tue Aug 27 18:23:17 UTC 2019
On Mon, Aug 19, 2019 at 12:22:30PM -0500, Mark Hatle wrote:
> On 8/19/19 11:49 AM, Otavio Salvador wrote:
> > On Mon, Aug 19, 2019 at 1:48 PM Mark Hatle <mark.hatle at windriver.com> wrote:
> >> On 8/19/19 11:27 AM, Ross Burton wrote:
> >>> On 19/08/2019 11:37, Otavio Salvador wrote:
> >>>> On Sun, Aug 18, 2019 at 11:00 PM Kang Kai <Kai.Kang at windriver.com> wrote:
> >>>>> On 2019/8/18 上午3:27, Otavio Salvador wrote:
> >>>>> Would you like to give more detailed steps how the error occurred? And
> >>>>> what's extra settings in you local.conf please?
> >>>>
> >>>> I had sysvinit and systemd enabled.
> >>>
> >>> The autobuilder should be exercising this, so when it's fixed we need to
> >>> ensure the AB covers the case Otavio is hitting.
> >>
> >> If they are both enabled what is the expected functionality of the system?
> >> (which is init, etc...)
> >
> > The image needs to pick the desired ones.
>
> Ok.. so build both packages.. and then come install time, one or the other must
> be in the PACKAGE_INSTALL (however the image chooses to do this), and the other
> should NOT be included.
>
> This could be problematic, as I've seen recipes build with systemd dependencies
> in them. (Both initscripts as well as system health monitoring and such.. That
> would conflict with a sysvinit [or systemd] if configured for the other.)
> Someone may need to review dependencies in the system looking for things that
> depend on systemd... or have a systemd/sysvinit 'conflict' of some kind. (I've
> not attempted such an audit before.)
>...
Dependencies on libsystemd are not a problem in sysvinit systems,
assuming there is proper runtime fallback code.
Shipping both an init script and a systemd unit file is also not a problem.
In Debian you can just switch the init system and reboot and it works,
with all the other binary packages unchanged.
> --Mark
cu
Adrian
BTW: There are problems like how to make GNOME happy without systemd,
and there is a lack of people in Debian working actively on the
sysvinit side. But on a high level the problems you were worried
about are "works in Debian".
More information about the Openembedded-core
mailing list