Rich Freeman <ri...@gentoo.org> wrote:
> Just a few clarifications below.
>
> One thing this discussion is missing is any mention of BIOS / EFI.
> Most of the discussion below seems most relevant to a legacy BIOS
> installation. Many specialized Gentoo install docs, like mdadm+lvm,
> don't really make mention of EFI, or other more recent developments.
> Now that all the docs are on the wiki I'd strongly encourage anybody
> with an interest to improve them. Many seasoned Gentoo users barely
> reference the documentation these days and I think that is part of why
> they've become a bit dated.
>
> A few of the topics below are somewhat controversial, particularly on
> this list. I tried to stick to the facts and indicate where there is
> a difference of opinion. I'd prefer not to rehash all the various
> debates...
>
> On Sun, May 17, 2015 at 9:24 AM, Peter Humphrey <pe...@prh.myzen.co.uk> wrote:
> > On Sunday 17 May 2015 12:48:58 Nuno Magalhães wrote:
> >
> >> (Later i want to get rid of systemd-udev and use eudev instead.)
> >
> > I use openrc, not systemd. It still works well and has less complication -
> > and
> > less typing!
>
> Most people using openrc are also using systemd-udev (and there is a
> good chance you do too). The latter was previously named udev and
> long predates what most people call systemd. Eudev is a fork of udev,
> which comes from after it came under the systemd umbrella, but before
> the name change and a number of changes that were controversial. I
> believe they try to incorporate many of the patches from systemd-udev
> but some default behaviors are different.
>
> In any case, I just wanted to clarify that systemd-udev is not the
> "systemd" you're probably thinking of. In particular, it doesn't
> replace openrc or sysvinit. Systemd-udev largely is concerned with
> populating /dev, and running initialization of hardware when it is
> detected, based on a configurable set of rules.
>
> >
> >> I intend to use XFS for /. Incidentally, if i later decide to "fork
> >> out" /usr (or some other subdirectory) into it's own LV, is it "just"
> >> a mater of copying its contents and updating /etc/fstab? Or should i
> >> just do it now and expand the LVs if later required (especially if i
> >> want to use different filesystems)?
> >
> > I can't help you with XFS. I know that ext4 in an LV in a VG in a PV on
> > RAID1
> > works reliably, even though it does look complex when I write it like that.
>
> As far as LVM and xfs themselves go, you can do what you propose.
>
> However, Gentoo QA policy is that it is expected that /usr is mounted
> early in boot. Various tools can break if it is not. Typically this
> is the responsibility of an initramfs, however you can also use
> scripts that run early during initialization from / which mount it.
> If you just stick /usr in fstab and rely on openrc to mount it for you
> normally, you may or may not have problems.
>
> It has been a long time since I actually used such a system in this
> manner with Gentoo, but the last I saw discussion on it most who used
> this configuration found it usually worked fine, unless you were using
> something like a bluetooth keyboard or other key system component that
> required a lot of userspace tooling to make work. However, as a
> matter of policy you're on your own if you choose not to mount /usr
> early during boot in some way.
>
> The reason it is not supported is that with the rise of things like
> bluetooth the list of dependencies possibly required during early boot
> has grown to the point where we'd end up not even having a /usr before
> long. My sense is that for the most part most maintainers tend to
> respect the traditional definition of / and /usr on Gentoo, and thus
> you can often get away with doing things the traditional way.
> However, the policy does allow us to end debates over things like udev
> rules invoking some userspace tool in /usr and such. Some packages
> more strongly depend on /usr being installed in early boot, and there
> have been suggestions (but nothing concrete) that Gentoo consider
> supporting the /usr-move that other distros have embraced (and that
> would basically get rid of /lib, /bin, and so on).
>
> >
> > Again, legacy grub here. But if you're using an initramfs, from what I've
> > seen
> > you don't need to specify metadata 0.90.
>
> I used to use grub legacy and kernel RAID auto-assembly. As a result
> I was using metadata 0.90.
>
> I found this configuration problematic on rare occasions. There is a
> reason that mdadm changed the metadata, and why most distros don't do
> it this way. (more below)
>
> >
> > Damn. I've just checked and something has renamed my /dev/md7 to
> > /dev/md127. Again. It's just too bad. I shall have to stop it when I get a
> > quiet moment and reassemble it into /dev/md7. Actually, I know what caused
> > it but I didn't notice at the time.
>
> And this was one of the configuration problems I ran into on rare
> occasion. Often booting from a rescue CD or such caused something
> like this to happen.
>
> One of the advantages of using an initramfs is that they can be a lot
> smarter about finding your partitions. You can identify them by UUID
> or label, and not care as much if mdadm or the kernel renames your
> device nodes.
>
> I'd seriously take a look at dracut, though I don't know if it works
> with eudev. It certainly should support openrc, and I know that it
> did back when I was running openrc. It can also mount /usr for you,
> and in fact it should automatically do so. It also respects your
> fstab - it uses its internal logic and the kernel boot line to
> initially find filesystems, but then it reads your /etc/fstab and
> remounts everything as you define it there just in case something has
> changed since the last time you built the initramfs/etc. You can
> define your own modules for it which makes it reasonably easy to get
> it to do anything at all during early boot, and it doesn't require
> anything to be built static (it finds required shared objects anywhere
> on the filesystem and includes them in the initramfs). It can also
> give you a rescue shell if something goes wrong, and depending on your
> settings you can make that rescue shell reasonably well-featured
> (using either dash or bash as you prefer inside, and I imagine you
> could tell it to install the other on the side). A while ago I needed
> to run some btrfs tools that aren't in dracut by default and it was
> trivial to tell dracut to include them, and I forced a shell on next
> boot which gave me the latest tools and kernel without having to build
> a rescue CD with them, and a bash shell to run them from.
>
> It certainly isn't necessary to use an initramfs to use Gentoo, and I
> used to be among the more minimalist crowd that avoided them.
> However, once I took the time to examine dracut it went from being a
> blob that looked unnecessary to a tool that is often useful.
Last time I tried to use dracut with openrc, it failed, I can't remember
exactly what happened, I think udev did hang, but its been a while since
this happened. Dracut uses systemd internally, so maybe this is part of
the problem.
--
Your life is like a penny. You're going to lose it. The question is:
How do
you spend it?
John Covici
cov...@ccs.covici.com
