On Fri, 7 Feb 2020 at 09:28, Zbigniew Jędrzejewski-Szmek <zbyszek@xxxxxxxxx> wrote:
On Fri, Feb 07, 2020 at 08:37:05AM -0500, Neal Gompa wrote:
> On Fri, Feb 7, 2020 at 8:27 AM Zbigniew Jędrzejewski-Szmek
> <zbyszek@xxxxxxxxx> wrote:
> >
> > On Fri, Feb 07, 2020 at 12:09:37PM +0000, Richard W.M. Jones wrote:
> > >
> > > I'd strongly suggest you look at supermin before going any further.
> >
> > Supermin is an interesting project, but at this point we're not
> > looking for a tool to craft the image. We're still at an earlier stage
> > of changing how we do some rpm packages so that it is easy to do an
> > installation that is small without custom logic.
> >
> > The idea is that the initramfs, once you include lvm, md, dm, encryption,
> > networking, clevis, emergency utilities, scsi, iscsi, raid in various
> > flavours, some graphics drivers, usb, bluetooth, etc, is a lot like a
> > the full distribution. Things would be a lot easier if we could just
> > use the same tools and daemons from the same packages as in the
> > host. Supermin (and other tools) have support for creating something
> > custom and small, and here the goal is the opposite: to do "standard".
> >
> > Fedora currently uses dracut, and the problem is that nobody has a
> > good grasp on what goes in dracut. There's a lot of custom logic and
> > custom code and reimplementation of things, and people who deal with
> > the same functionality in the host system don't necessarily know what
> > happens in the initramfs. In the past such a setup made sense, but now
> > it seems that the tradeoffs are different.
> >
>
> I'm genuinely surprised by this. [...] The main problem with
> dracut is the same problem with all initramfs generators: it has to
> prepare an environment that works in the worst circumstances. It's
> compounded by the restriction that everything is written in shell, and
> POSIX shell at that (because Debian).
Well, look at it from the other side: the host machine also needs to
work in the same circumstances, since we need things to work also after
we switch to the host. And the actual code that we use in the host
— the libraries, the binaries, configuration — is the same, since we
don't do different compilations for the initramfs.
In the host we have switched away from the shell for machine boot, but
for some reason we still keep it for the initramfs, along with a large
amount of custom logic.
> The dracut package, while very large,
> has a pretty understandable framework model.
Sure, it *can* be understood, but should it? Do we gain anything by
having the initramfs so different from the host? The initial
My problem here is that we continually get told we are full of Not Invented Here(NIH) solutions by people in other communities and we should be doing something Debian and different communities use..
And here is one of the places we do agree with other distributions, but it is now a "bah its crap and you can do something better that works with your tools." We knew that years ago and we have known that time and time again. The issue is that we end up with another 'NIH' tool which our limited manpower are going to know and as soon as XYZ developer walks off to $ABC startup we have an unbootable and unmaintained mess. AKA where we were before dracut.
If we are hell bent on repeating history, please let us try to learn from it first.
Please allow me to weigh in here, whilst I straddle this fence since I have a vested stake here, though I only have $0.02 to vote with just like anyone else.
On one hand, I like the idea being proposed. I mean, it
shouldn't be that different of an environment than what you're
trying to boot into. The whole process has a a lot of complexity
that seems wasteful in some cases but affords flexibility in
others. Nobody likes waiting for dracut to do its thing. I've
literally done so hundreds of times in the last few weeks as I
prepare for an in-service, in-place upgrade of several hundred
nodes running a custom Fedora Live spin to a newer custom Fedora
Live spin. I've done so successfully shepherding them from Fedora
10 to 25 and am about to jump them all to F29. The magic all
happens via a custom initrd that does a switcheraroo of the LiveOS
and syslinux directories. So, one reboot into the magic initrd to
do the switch and another reboot from there into the new distro.
(I should note, that the hundreds of dracut runs are its fault,
but instead for dev work related to how the new payload is
managed. Nonetheless, I've done a lot of waiting in testing due
to dracut runs.)
On the other hand, I remember the madness of trying to maintain
local patches to work with the contemporary initrd generation
process that predated dracut and it was awful. Every iteration
required tweaks. Dracut was another tweak, a big one. But you
know what? Once my process adapted to dracut, it really hasn't
needed to change since -- thanks to a beautiful plugin setup.
Adoption wasn't too hard either thanks to easy-to-read shell
script (some of the best I've seen).
Can this be improved? I'm sure it can be. I'm genuinely curious what dracut does that an rpm couldn't ship. I'd also hate to lose the versatility that we have at the present. I've been very successful with what Fedora has offered, but I've always been scared of autonomous nodes building/booting their own special initrd's and hoping all just works.
I guess I just split my $0.02 in half.
_______________________________________________ devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/ List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines List Archives: https://lists.fedoraproject.org/archives/list/devel@xxxxxxxxxxxxxxxxxxxxxxx
