Bug#574336: pkgsel: Only run update-initramfs once during installation?

[Petter Reinholdtsen]

Package: pkgsel
Version: 0.25
Tags:patch

When checking the Ubuntu changes to pkgsel, I noticed this change:

 - Divert update-initramfs and call it just once at the end (with
   bind-mounted /dev, /proc, and /sys, and a preceding update-dev).

This seem like a good optimization for the installation time to bring
over to Debian, making sure update-initramfs is only executed twice
during the installation instead of several times more (once when the
kernel is installed by base-insatller, and once after pkgsel is done).
Is there any reason not to include it?

This patch relative to the current svn version would implement it.  It
is ripped directly from the Ubuntu patch.

Index: debian/postinst
===
--- debian/postinst (revision 62670)
+++ debian/postinst (working copy)
@@ -6,7 +6,7 @@
 export LANG=C
 
 REMOUNT_CD=
-DIVERTS='/usr/bin/scrollkeeper-update /usr/bin/scrollkeeper-rebuilddb 
/usr/bin/fc-cache'
+DIVERTS='/usr/bin/scrollkeeper-update /usr/bin/scrollkeeper-rebuilddb 
/usr/bin/fc-cache /usr/sbin/update-initramfs'
 
 log() {
logger -t pkgsel $@
@@ -163,7 +163,22 @@
chroot /target fc-cache -f -v /target/var/log/fontconfig.log 21 \
|| true
 fi
+if [ -x /target/usr/sbin/update-initramfs ]; then
+   # get UUIDs for any devices formatted after partitioning
+   update-dev
 
+   # make UUIDs etc. available in the target system
+   mount -o bind /dev /target/dev
+   mount -o bind /sys /target/sys
+   mount -o bind /proc /target/proc
+   
+   chroot /target /usr/sbin/update-initramfs -u || true
+   
+   umount /target/proc
+   umount /target/sys
+   umount /target/dev
+fi
+
 db_progress STEP 30
 db_progress STOP
 
Happy hacking,
-- 
Petter Reinholdtsen



-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org

Bug#574336: pkgsel: Only run update-initramfs once during installation?

[Frans Pop]

On Wednesday 17 March 2010, Petter Reinholdtsen wrote:
 This seem like a good optimization for the installation time to bring
 over to Debian, making sure update-initramfs is only executed twice
 during the installation instead of several times more

In exactly which scenario would it be executed more than twice?

I don't much like the patch either. Why doesn't it just use in-target to 
run update-initramfs?



-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org

Bug#574336: pkgsel: Only run update-initramfs once during installation?

[Petter Reinholdtsen]

[Frans Pop]
 In exactly which scenario would it be executed more than twice?

I expect it to run the triggers several times, once for each block of
packages passed from apt to dpkg, but have not verified that this is
the case.  I had a look in the installation log for Debian Edu based
on Squeeze, and here it is called four times:

r...@tjener:~#  grep -C1 update-init /var/log/installer/syslog
Mar 14 06:03:00 in-target: Setter opp initramfs-tools (0.93.4) ...
Mar 14 06:03:00 in-target: update-initramfs: deferring update (trigger 
activated)
Mar 14 06:03:00 in-target: Behandler utløsere for initramfs-tools ...
--
Mar 14 06:03:12 in-target: Running depmod.
Mar 14 06:03:12 in-target: Running update-initramfs.
Mar 14 06:03:12 in-target: update-initramfs: Generating 
/boot/initrd.img-2.6.32-trunk-686
Mar 14 06:03:22 in-target: Examining /etc/kernel/postinst.d.
--
Mar 14 07:11:44 in-target: update-alternatives: using 
/usr/lib/usplash/debian-edu-usplash.so to provide 
/usr/lib/usplash/usplash-artwork.so (usplash-artwork.so) in auto mode.
Mar 14 07:11:44 in-target: update-initramfs: deferring update (trigger 
activated)
Mar 14 07:11:44 in-target: Setter opp debian-edu-doc-en 
(1.3~20100223~5.0.4+edu0) ...
--
Mar 14 07:13:51 in-target: Setter opp usplash (0.5.19-3) ...
Mar 14 07:13:52 in-target: update-initramfs: deferring update (trigger 
activated)
Mar 14 07:13:52 in-target: Setter opp util-linux-locales (2.16.2-0) ...
--
Mar 14 07:15:18 in-target: Behandler utløsere for initramfs-tools ...
Mar 14 07:15:18 in-target: update-initramfs: Generating 
/boot/initrd.img-2.6.32-trunk-686
Mar 14 07:15:28 debian-edu-fsautoresize: overfull /usr, resizing to 6596198
--
Mar 14 07:46:15 in-target: KILLALL=false
Mar 14 07:46:15 in-target: update-initramfs: deferring update (trigger 
activated)
Mar 14 07:46:16 in-target: Setting up libklibc (1.5.15-1) ...
--
Mar 14 07:46:16 in-target: Setting up initramfs-tools (0.93.4) ...
Mar 14 07:46:16 in-target: update-initramfs: deferring update (trigger 
activated)
Mar 14 07:46:16 in-target: Setting up syslinux-common (2:3.85+dfsg-1) ...
--
Mar 14 07:46:34 in-target: Setting up dmsetup (2:1.02.39-1) ...
Mar 14 07:46:34 in-target: update-initramfs: deferring update (trigger 
activated)
Mar 14 07:46:34 in-target: Setting up cryptsetup (2:1.1.0~rc2-1) ...
Mar 14 07:46:34 in-target: update-initramfs: deferring update (trigger 
activated)
Mar 14 07:46:34 in-target: Setting up libavahi-common-data (0.6.25-3) ...
--
Mar 14 07:47:36 debian-edu-fsautoresize: overfull /var, resizing to 3216998
Mar 14 07:47:37 in-target: Running update-initramfs.
Mar 14 07:47:37 in-target: update-initramfs: Generating 
/boot/initrd.img-2.6.32-trunk-486
Mar 14 07:47:46 in-target: cryptsetup: WARNING: could not determine root device 
from /etc/fstab
--
Mar 14 07:48:03 in-target: update-alternatives: using 
/usr/lib/usplash/debian-edu-usplash.so to provide 
/usr/lib/usplash/usplash-artwork.so (usplash-artwork.so) in auto mode.
Mar 14 07:48:03 in-target: update-initramfs: deferring update (trigger 
activated)
Mar 14 07:48:03 in-target: Setting up firmware-linux-free (2.6.32-5) ...
--
Mar 14 07:48:08 in-target: Setting up usplash (0.5.19-3) ...
Mar 14 07:48:08 in-target: update-initramfs: deferring update (trigger 
activated)
Mar 14 07:48:09 in-target: Setting up tasksel-data (2.81) ...
--
Mar 14 07:48:09 in-target: Processing triggers for initramfs-tools ...
Mar 14 07:48:09 in-target: update-initramfs: Generating 
/boot/initrd.img-2.6.32-trunk-486
Mar 14 07:48:18 in-target: cryptsetup: WARNING: could not determine root device 
from /etc/fstab
r...@tjener:~#  grep -C1 update-init /var/log/installer/syslog|grep Gener
Mar 14 06:03:12 in-target: update-initramfs: Generating 
/boot/initrd.img-2.6.32-trunk-686
Mar 14 07:15:18 in-target: update-initramfs: Generating 
/boot/initrd.img-2.6.32-trunk-686
Mar 14 07:47:37 in-target: update-initramfs: Generating 
/boot/initrd.img-2.6.32-trunk-486
Mar 14 07:48:09 in-target: update-initramfs: Generating 
/boot/initrd.img-2.6.32-trunk-486
r...@tjener:~#

It is possible some of these are from setting up the LTSP chroot
environment.

 I don't much like the patch either. Why doesn't it just use
 in-target to run update-initramfs?

I did wonder about that too.  I expected in-target to handle /proc/
and friends, but do not know why Ubuntu choose to do this outside
in-target and did not test if they are needed or not.  Decided to pass
on the patch untouched for review. :)

Happy hacking,
-- 
Petter Reinholdtsen



-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org

Bug#574336: pkgsel: Only run update-initramfs once during installation?

[Frans Pop]

On Wednesday 17 March 2010, Petter Reinholdtsen wrote:
 It is possible some of these are from setting up the LTSP chroot
 environment.

Something looks very broken in this example.

Why is initramfs-tools getting set up twice for the same version?

Why is dmsetup getting installed if it is not being used during 
partitioning (in which case it should already have been installed _before_ 
initramfs-tools gets installed the first time)?

After that the only remaining cause for an extra run is usplash.
If debian-edu wants that installed by default, then the correct solution 
IMO is to include it on the first CD and ensure it gets installed during 
post-base-installer.d and not during pkgsel.

The problem here IMO is a badly structured installation, not a missing 
feature in pkgsel.

Cheers,
FJP



-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org

Bug#574336: pkgsel: Only run update-initramfs once during installation?

[Petter Reinholdtsen]

[Frans Pop]
 Something looks very broken in this example.
 
 Why is initramfs-tools getting set up twice for the same version?

No idea.  This what made me suspect the cause might be the LTSP
chroot.

 Why is dmsetup getting installed if it is not being used during
 partitioning (in which case it should already have been installed
 _before_ initramfs-tools gets installed the first time)?

It is a dependency of lvm2.  I asked 'aptitude why dmsetup' on the
machine in question. :)

 After that the only remaining cause for an extra run is usplash.  If
 debian-edu wants that installed by default, then the correct
 solution IMO is to include it on the first CD and ensure it gets
 installed during post-base-installer.d and not during pkgsel.

There is not enouch room on the CD for that. :)

 The problem here IMO is a badly structured installation, not a
 missing feature in pkgsel.

Oh, well.  My goal for this request was a faster installation.  I
assume we share that goal, and do not see the point of arguing further
for making sure update-initramfs is called as few times as possible.

Happy hacking,
-- 
Petter Reinholdtsen



-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org

Bug#574336: pkgsel: Only run update-initramfs once during installation?

[Frans Pop]

On Wednesday 17 March 2010, Petter Reinholdtsen wrote:
 No idea.  This what made me suspect the cause might be the LTSP
 chroot.

Ah. I failed to really comprehend that earlier.
Yes, I guess that could be and that would of course make it separate from 
base system installation 

  Why is dmsetup getting installed if it is not being used during
  partitioning (in which case it should already have been installed
  _before_ initramfs-tools gets installed the first time)?

 It is a dependency of lvm2.  I asked 'aptitude why dmsetup' on the
 machine in question. :)

The same goes for LVM2 though. That should really be installed as a result 
of partitioning, not as an extra package. Or is that installed in the LTSP 
chroot as well?
If it is the LTSP chroot that is the cause, then I think pkgsel really is 
the wrong place to address this.

 Oh, well.  My goal for this request was a faster installation.  I
 assume we share that goal, and do not see the point of arguing further
 for making sure update-initramfs is called as few times as possible.

Sure. But that does not mean we should blindly complicate D-I code without 
checking the validity of the use case.

P.S. I've been wondering if we shouldn't consider postponing kernel 
installation until after APT is configured. In some ways it is more 
logical to group it with bootloader installation than with base system 
installation. It could also make sense for firmware installation (see my 
comments in #574158). Not something to change lightly though. Added 
complication is consistent handling of Recommends.



--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org