Bug#940028: launching d-i on multiple consoles doesn't play well with preseeding

[Cyril Brulebois]

Hi Sergio,

Sergio Gelato  (2019-09-11):
> This seems to be new behaviour since stretch, introduced in version 1.128 of
> package rootskel.

From the rest of your mail I'm assuming you meant buster here.

> I suspect that the two debian-installer instances are racing each
> other. To confirm this, I unpacked initrd.gz, made a simple change to
> sbin/reopen-console (adding inittab entries only for cons in
> $preferred instead of for cons in $consoles), repacked, and run an
> installation again with much better results (no more questions from
> localechooser).
> 
> I do appreciate that for interactive d-i use it's nice to have it show
> up on all consoles; but not at the expense of breaking fully automated
> installation.

Agreed. You're not the only one reporting such issues, and I'd really
like feature authors to fix the regression they introduced. That's not
the first time steal-ctty appears in bug report.


Cheers,
-- 
Cyril Brulebois (k...@debian.org)
D-I release manager -- Release team member -- Freelance Consultant


signature.asc
Description: PGP signature

Bug#940028: launching d-i on multiple consoles doesn't play well with preseeding

[Sergio Gelato]

Package: rootskel
Version: 1.131
Severity: important

I like to create Xen domU virtual machines using debian-installer, with
preseeding for a no-questions-asked experience. This works well with stretch,
but when I try to use buster images (from 
debian/dists/buster/main/installer-amd64/current/images/netboot/xen/ in the 
public repository) I find that I am
being asked some questions (by localechooser) despite the preseeding. Moreover,
the exact questions asked vary from run to run even if I make no changes to
my setup. Occasionally I even get netcfg failures.

>From the installer log I see that two instances of debian-installer are being
run in parallel, one for each of the consoles hvc0 and tty0:

Sep 10 10:02:13 reopen-console: Looking at console hvc0 from /proc/consoles
Sep 10 10:02:13 reopen-console:Adding hvc0 to consoles list
Sep 10 10:02:13 reopen-console:hvc0 is preferred
Sep 10 10:02:13 reopen-console: Looking at console tty0 from /proc/consoles
Sep 10 10:02:13 reopen-console:Adding tty0 to consoles list
Sep 10 10:02:13 reopen-console: Adding inittab entry for hvc0
Sep 10 10:02:13 reopen-console: Adding inittab entry for tty0
Sep 10 10:02:13 reopen-console: Restarting init to start d-i on the consoles we 
found
Sep 10 10:02:13 init: reloading /etc/inittab
Sep 10 10:02:13 init: starting pid 208, tty '/dev/tty4': '/usr/bin/tail -f 
/var/log/syslog'
Sep 10 10:02:13 init: starting pid 209, tty '/dev/hvc0': 
'/sbin/debian-installer'
Sep 10 10:02:13 init: starting pid 210, tty '/dev/tty0': 
'/sbin/debian-installer'

This seems to be new behaviour since stretch, introduced in version 1.128 of
package rootskel. I suspect that the two debian-installer instances are
racing each other. To confirm this, I unpacked initrd.gz, made a simple change
to sbin/reopen-console (adding inittab entries only for cons in $preferred
instead of for cons in $consoles), repacked, and run an installation again
with much better results (no more questions from localechooser).

I do appreciate that for interactive d-i use it's nice to have it show up on
all consoles; but not at the expense of breaking fully automated installation.