--- Begin Message ---
- To: submit@bugs.debian.org
- Subject: Bugs in various documents, bugs with partitioning, bugs with LVM, bugs with mkinitrd.
- From: "Dr. Andreas Krüger" <andreas.krueger@dv-ratio.com>
- Date: Wed, 21 Apr 2004 18:18:08 +0200
- Message-id: <40869EC0.4010607@dv-ratio.com>
Package: installation-reports
Version: unavailable; reported 2004-04-21
My experiences with the Sarge installer.
I want the following setup:
/dev/hda1, about 100 MByte, ext2, holds grub and kernel and initrd
/dev/hda2, several gig, LVM, holds /, and swap, and all the rest.
I found the installation howto
http://http.us.debian.org/debian/dists/testing/main/installer-i386/current/doc/INSTALLATION-HOWTO
I got
http://people.debian.org/cdimage/testing/sarge_d-i/i386/current/sarge-i386-businesscard.iso
(dated 19-Apr-2004).
Also, I don't want to go through the trouble of burning a CD.
So, while the CD image still downloads (internet connection to people.debian.org
was slow yesterday), I boot the computer I want to install, with a Knoppix-CD.
I prepare partitions /dev/hda1 and /dev/hda2 the size I want. I do mke2fs
/dev/hda1, install grub on boot sector /dev/hda, and grub support files on
/dev/hda1.
The installation howto tells me I need hd-media/initrd.gz and hd-media/vmlinuz.
Where do I find these? I count this as a bug of the installation howto (as
above): It should give URLs for those files.
Finally, download is through and MD5
2bf3e5a7cdc991e9fc3362cc8d8cd42e *sarge-i386-businesscard.iso
comes out fine.
I still don't know where to get hd-media/initrd.gz and hd-media/vmlinuz.
On sarge-i386-businesscard.iso itself, I find
doc/install/manual/en/ch04s02.html#where-files
Yes! This tells me where those files are, namely
http://http.us.debian.org/debian/dists/sarge/main/disks-i386/current/images/hd-media/
No! They aren't there. This is a bug in ch04s02.html on the iso image: The link
is broken. After some searching, I find those files
http://http.us.debian.org/debian/dists/sarge/main/installer-i386/current/images/hd-media/
I copy those onto /dev/hda1 of that computer that's still under Knoppix. I
reboot it and do as directed in the howto, booting into the iso image through grub.
No, says the kernel. That ramdisk is 8731k, not 8192 as the installation howto
would have me believe. So I count this as a bug in the installation howto: The
ramdisk size given is not correct.
Booting through grub, new try, changing that number. Fine, I'm in the ususal
Debian installation system. Trying the German translation.
DHCP configuration works out fine. I like that.
HTTP proxy? Well, thank you for asking. Given a choice, I would rather like to
use the apt-proxy installation I have. So this is a wishlist bug: I would like
to have an option to use an existing apt-proxy in my local network.
Partitioning. Wow, it DOES support LVM! I like that!
I try to tell it to leave the /dev/hda1 alone, accept what's there, and mount it
as /boot. This does not work. I get a mysterious error about "parted" not being
able to resize the ext2 partition. I had not intended to resize this partition.
Just use as is.
The error message does not relate to what I belive the actual problem is: The
partition /dev/hda1 is in use. I have the CD-image on /dev/hda1, on which the
loop device lives. Appearently, the installation software tries to un-mount or
re-mount it or whatever? This cannot work.
So, I call this a wishlist bug: I want to be able to use a Linux partition to
hold the CD image sarge-i386-businesscard.iso that's being actively used for the
installation, and at the same time I want to be able to add this partition,
keeping all stuff that's on there, as some mount point of my choosing into the
installation that's being generated.
So I try to leave /dev/hda1 without a mount point for the new system-to-be.
Which gets me into the installation of the base system all right.
As far as the installation is concerned, I now ONLY have LVM partitions. Is
that possible? Is it possible to boot, using grub, into such a setup? Does grub
support itself living on LVM? I doubt it...
Call this another bug or wishlist bug (your choice): When the user partitions
"all on LVM", the installation system should not even try to invoke grub later.
(I'm not even all that sure calling mkinitrd makes a whole lot of sense in
such a situation.)
A simple cat /proc/cpuinfo reveals what CPU I have, but the installation
suggests a plain vanilla 386 kernel. I choose "kernel-image-2.6.3-686". That
runs into trouble:
"/usr/sbin/mkinitrd: constituent device /dev/vg/rootlv does not exist.
Failed to create initrd image."
So I erase the designated "/" partition and starting over with installing
packages on it, as the installation system recommends I should. This is weird:
This time around, I don't get to choose the kernel again, but the installer
simply picks kernel-image-2.4.25-1-386.
This does not help: With kernel 2.4.25, I still run into the very same problem I
saw earlier, with the 2.6.3 - kernel: Cannot build mkinitrd.
So this is yet another bug: Of course, mkinitrd can be talked into producing an
initrd that's preparing the kernel for / on LVM. But the installation system
does not do the required talking.
Ok. I'm going to fix this manually, with Knoppix. So I choose to say,
"continue without boot loader".
Oooops - the installation process seems to start all over again. This is not
what I need! Yet another bug: I cannot continue when the installation system
thinks there's no kernel/initrd yet.
Ok, lessons learned. Rebooting, starting all over, from scratch.
I have 128 MB RAM on this machine. So I want to move the cd image into tmpfs,
thus untie my hands in using the hard disk. Then, for the time being, use
/dev/hda1 for / . For what I know at this point, this just might work.
When I'm asked for the systems name, I don't answer yet. I sneakingly open the
second console, mount a 30 MB tmpfs somewhere, copying the iso - image from hard
disk to tmpfs, umounting /cdrom from loop on the .iso - file that lives on
/dev/hda1. Hey, I'm lucky for a change: This file system happens to be not in
use at this point. Good. I mount /cdrom (through loop, again) to the copy of
the iso - image that now lives on tmpfs. Now I can umount /dev/hda1. Deep
sigh. The entire hard disk is not in use any more. This should help. Back to
first console. Here is your host name.
A short time later, I'm glad we not only have an apt-proxy, but also a plain
http proxy cache in our LAN.
The new approach does get me around the various partion problems, but not around
the mkinitrd problem, as I had hoped. Sigh.
So there is a bug when I want "/" on LVM. I'm still unable to mkinitrd.
Ok, try something else, then. Forget LVM, for the time being. Fortunately,
/dev/hda1 is 93 MB. This should be enough to hold "/" of a very basic system.
So I tell the installation system to make /dev/hda1 my new root system. But
leave all data on /dev/hda1 alone, please.
Nope. This throws me into some "parted" - loop again.
Sigh. So, instead, kill /dev/hda1. After all, I do have those files both in
RAM and on a different computer. I allow the installation system to start
partitioning anew, create a new file system on /dev/hda1. I ask to have swap on
LVM, but no longer "/".
No, sorry. The 93 MB that should be plenty for a few kernels, grub, and initrd -
images, are no longer enough to hold a Debian base installation. In the days of
bo or so, I have been running a complete Debian with XWindows and what have you
on 200 MB. Those days are gone, it seems. I get an "out of space" error.
This is yet another wishlist bug: Not to bring old days back. But the system
should know how much space is needed. At least, as long as only mount points
from the list of suggestions are used. If I'm trying to install on too little
space, I should be warned early.
But I still have the installation CD in RAM. So go back to partitioning.
Now I get a warning, that the kernel was unable to re-read the partition table
on /dev/ide/host/bus0/target0/lun0/disc . That's the end of this attempt, it
seems. Back to Knoppix, start all over again?
Just to see what happens, I first try to partition manually, using fdisk. On
"w", fdisk tells me the problem comes from the device being in use. I have a
look at ps fax and see that there is several programms running with "part" in
their name. I kill all these. Let's see what happens now.
No, that did not help.
Yesterday: Deep sigh. Boot into Knoppix.
This is yet another bug: When I goof the partitioning phase, I expect to be able
to start over again with it, without starting ALL over again (reboot).
Another bug I noticed (which may well be the source of the bug I just
mentioned): When I delete the underlying partition for LVM, the individual
logical volumes should disappear. They don't. Once I initialized the volume
manager, I'm unable to completely undue the effects.
Afterthought, a day later: At that point, I might have swap still connected to
some logical volum?
Yesterday, a while later: No, even almost 200 MB isn't enough for a Debian base
these days. So I decide to use LVM for /usr .
Finally, this works. I'm through the initial stage of my system installation.
Debian installs grub. Reboot. I see a grub prompt.
This took longer than I had hoped it would. It is late now. I don't boot into
the newly installed system. I switch off and go home.
That way yesterday. Today, I would appreciate if someone who's more
knowledgeable than I'm regarding the different parts that make up the
installation system, could rip this apart and distribute my bugs to the
individual packages where they belong.
Regards, and thank you for constructing fine software
Andreas
--
Dr. Andreas Krüger, andreas.krueger@dv-ratio.com
DV-RATIO Nordwest GmbH, Tel.: +49 211 577 996-0, Fax: +49 211 559 1617
Leostraße 31, 40545 Düsseldorf, Germany
--- End Message ---