Re: Re: Minimal snapshots for cooker available.

Jeremiah Summers <[email protected]> Fri, 20 Jan 2012 15:14:16 -0700
Newsgroups gmane.linux.mandrake.cooker.devel
Message-ID <CAJeqQ0EutesaYSU_psByfzRVyzjx_aKSsL9-W_BU8Bb_AkeXpw@mail.gmail.com>
On Fri, Jan 20, 2012 at 2:07 PM, Matthew Dawkins <[email protected]> wrote:
>
>
> On Fri, Jan 20, 2012 at 11:48 AM, Per Øyvind Karlsen
> <[email protected]> wrote:
>>
>> Den 19:14 20. januar 2012 skrev Matthew Dawkins <[email protected]>
>> følgende:
>> >
>> >
>> > On Thu, Jan 19, 2012 at 1:11 PM, Franck Bui <[email protected]>
>> > wrote:
>> >>
>> >> Matthew Dawkins <[email protected]> writes:
>> >>
>> >> [...]
>> >>
>> >> >
>> >> > What kmod is that? i have been checking the kernels ever since you
>> >> > mentioned it to me.
>> >>
>> >> I think I've already described how to get/use it somewhere else...
>> >>
>> >> > My main problem ATM are building solid initrd imgs and the unioning
>> >> > fs
>> >> > needed in the livecd. Currently the kernels are requiring mkinitrd
>> >> > and
>> >> > not dracut.
>> >>
>> >> If you can detail further maybe I could have a look and help.
>> >>
>> >> > We could drastically clean up what is needed in the initrd if we
>> >> > started migrating to dracut and a static busybox.
>> >>
>> >> Perhaps, but my goal was not to have the smallest image, but a snapshot
>> >> of current cooker ready to use in various context.
>> >>
>> >> --
>> >>                Franck
>> >
>> >
>> >
>> > Well, maybe we can address one issue at a time.
>> Aren't you doing the exact opposite here..? ;p
>
> :P
>>
>> >
>> > What would it take to get dracut to replace mkinitd?
>> It's already replaced mkinitrd since a while ago (although for people
>> upgrading and
>> already using mkinitrd, this stayed as default).
>
> The kernels afaik from checking still require mkinitrd
> If dracut is working and there are no problems booting with initramfs
> images, then way not finally switch?
>>
>> > And I think dracut should prolly prefer the static busybox over the
>> > uclibc
>> > compile busybox.
>> dracut doesn't use busybox at all currently.
>>
>> The statically linked busybox package is anyways linked against
>> uClibc, and as you'd also
>> want to bundle ie. plymouth and other stuff that busybox doesn't
>> provide, it would be preferred
>> to use binaries dynamically linked against uClibc to save space.
>>
>> It's anyways not really a topic worth much discussion before someone
>> does the job of adding
>> support for busybox to dracut though (which would be nice as we then
>> could reduce initrd size
>> from > 32MB all the way down to 1 MB)..
>
> dracut just produced a ~7MB size initramfs for me...
> but yes busybox removes the need for udev, modprobe and most of the libs
> afaik
>>
>>
>> And it's rather off-topic for this thread.
>>
>> --
>> Regards,
>> Per Øyvind
>
>

The init for a Live System should role is to find the root drive and
mount it then do the swap. Dracut doesn't even need to be used for
this, I think this what Matt is talking about. He might be a little
confused over uClibc, which fedora and various of distributions use to
statically build busybox. Busybox however space wise seems to be a
interesting solution rather then pulling in various binaries and
needed libs from the system as I assume dracut does. Dracut works well
for a mkinitrd replacement outside of live environments but at least
in my experience is not efficient enough just yet to replace a
generated init with something like an all in one solution like
busybox, even if you choose not to use busybox's mdev for udev.