[01:29] <cjwatson> that's not surprising; if you randomly kill debconf, extra questions are the least of your worries
[01:30] <cjwatson> if it's happening about half the time, I'd love it if you could step through the installer until just before the end of the partitioner, then from tty2, 'udpkg -i /cdrom/pool/main/s/strace/strace_*.deb', 'ps ax' to find the main-menu process, and 'strace -f -o /tmp/foo.trace -s 1024 -tt -p <whatever the main-menu pid is>'
[01:31] <cjwatson> ctrl-c that if/after it hangs, 'anna-install openssh-client', and scp the trace out
[01:34] <cjwatson> this is a little reminiscent of http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=100616, but I don't think it's the same thing
[11:38] <jerom1> Hi Colin and all,
[11:38] <jerom1> I would wish to test your friday's patch by adding your modification in the initrd of my netboot. How can i find your fix ?
[01:51] <mark> cjwatson: alright I'll see what I can do
[01:51] <cjwatson> thanks
[01:53] <mark> I guess I could start an openssh server on the machine if I'm doing the install manually step by step anyway
[01:54] <cjwatson> possibly the fewer perturbations the better, though
[01:55] <mark> yes but I don't have physical access to most of our machines
[01:55] <mark> they are thousands of miles away ;)
[02:09] <mark> let me try it rightaway on an idle server in korea
[02:41] <mark> of course it doesn't happen if you WANT it to happen
[02:43] <cjwatson> a watched pot never boils
[02:44] <mark> maybe it doesn't help that the install goes a lot slower now because of a different mirror
[02:44] <cjwatson> that shouldn't make a difference; it downloads everything and then installs it all
[02:45] <mark> you never know
[03:00] <mark> aha
[03:00] <mark> now the install seems stalled on "Installed libisccfg1"
[03:00] <mark> so it's likely not a man-db issue but a debconf one?
[03:05] <cjwatson> could be anything
[03:06] <cjwatson> would need an strace to investigate
[03:07] <mark> trying to get you one now
[03:07] <mark> btw, once I used db_set in a postinst script in one of our own packages
[03:08] <mark> and that made everything stall as well
[03:08] <mark> might be the same thing
[03:08] <mark> but that was very reproducable
[03:11] <cjwatson> probably a different problem, such as causing debconf to exit before the db_set
[03:11] <cjwatson> db_set does work :-)
[03:12] <mark> server-expert installs at 9600bps = no fun :)
[03:27] <mark> ah there is no openssh-server udeb
[03:28] <mark> let's see if I can do without...
[03:41] <cjwatson> is too
[03:41] <cjwatson> but it comes unconfigured so is not really quite what you want
[03:41] <cjwatson> (it's in universe)
[03:55] <mark> I have found another way
[03:55] <mark> I start the strace in the ssh shell using the menu
[03:55] <mark> and have it log to /var/log/stall.trace
[03:55] <mark> then I start the internal debug logs webserver
[03:55] <mark> and go on with the install
[03:57] <mark> ok it seems stalled at "Installed cron"
[03:57] <mark> http://yf1014.yaseo.wikimedia.org/stall.trace
[04:01] <cjwatson> jerom1: 'svn di -r42868:42869 svn://svn.debian.org/svn/d-i/trunk/packages/debian-installer-utils/list-devices', applied to /bin/list-devices. If it breaks, you get to keep both pieces.
[04:01] <jerom1> collin : thanks
[04:01] <mark> 2136  14:56:57.569926 sendto(4, "<15>Nov 27 14:56:57 debconf: <-- 0 OK", 37, MSG_NOSIGNAL, NULL, 0) = 37
[04:01] <mark> 2136  14:56:57.570009 write(5, "0 OK", 4) = 4
[04:01] <mark> 2136  14:56:57.570056 write(5, "\n", 1) = 1
[04:01] <mark> 2136  14:56:57.570102 read(6,
[04:01] <cjwatson> (just the one l)
[04:01] <mark> are the last lines in the trace
[04:01] <cjwatson> that sounds like you're just tracing debconf
[04:02] <mark> isn't the main menu debconf?
[04:02] <cjwatson> but the hang might not be there
[04:02] <cjwatson> you're prejudging the outcome ...
[04:02] <mark> I am not
[04:02] <mark> you asked me to trace the main menu pid :)
[04:02] <cjwatson> with -f
[04:02] <mark> I did
[04:02] <cjwatson> and no, the main menu actually isn't debconf
[04:02] <cjwatson> it's "main-menu"
[04:02] <mark> ok
[04:03] <mark> I will start over
[04:03] <cjwatson> sorry, I thought I'd been clear
[04:03] <mark> maybe I didn't look very well
[04:03] <mark> np
[04:03] <mark> I now know how to do it
[04:03] <cjwatson> looking at the trace anyway, but it looks like it's the other side of the protocol that's hanging
[04:03] <mark> yes
[04:03] <cjwatson> debconf is just waiting for input, which is legitimate and normal
[04:04] <mark> indeed
[04:04] <cjwatson> could conceivably be apt or something at this rate
[04:04] <cjwatson> but we'll see
[04:08] <mark> I should put the strace in the early_command script... easier
[04:19] <mark> oh this is going to be a fun trace
[04:20] <cjwatson> (not likely to be around much for the rest of my afternoon, BTW)
[04:20] <mark> ok
[04:20] <mark> I'll just save it
[04:20] <cjwatson> nnn
[04:20] <cjwatson> (argh, bloody connection)
[04:21] <mark> gee, the installer is doing a lot of forks...
[05:00] <cjwatson> it does that
[05:00] <cjwatson> something to do with largely being written in shell ;-)
[05:01] <cjwatson> Doesn't help performance on hppa any, where fork() is really slow. It's not too bad on the other architectures.
[05:26] <mark> hehe
[05:26] <mark> my strace download is now one hour behind
[05:26] <mark> an hour to just install base ;)
[05:54] <mark> killed for out of memory
[05:55] <cjwatson> d'oh
[05:55] <mark> it was > 4 GB in size ;)
[05:55] <mark> this is not feasible
[05:55] <cjwatson> when did you start the trace?
[05:55] <mark> right before base install
[05:55] <cjwatson> hmm
[05:56] <cjwatson> is it feasible to dump the trace to disk instead of RAM?
[05:56] <mark> hmm
[05:56] <mark> maybe
[05:56] <mark> would have to be another disk than the installer is working on I guess
[05:56] <cjwatson> you'd have to run in expert mode until the end of the partitioner in order to give yourself time to set up the trace before base-installer starts, and then change debconf priority back to high to get into base-installer
[05:57] <cjwatson> no, if you set up the trace after the partitioner has finished working then you can safely write into /target/tmp/
[05:58] <mark> or I have it streamed out over the LAN
[06:09] <mark> I can do it after base-installer I guess, it doesn't happen during bae
[06:09] <mark> base
[06:11] <cjwatson> it doesn't?
[06:11] <cjwatson> oh, Ubuntu doesn't install man-db as part of base
[06:11] <cjwatson> I totally forgot about that ;-)
[06:11] <mark> :)
[06:11] <cjwatson> hi, can you tell I'm the man-db maintainer, a Debian installer developer, and the Ubuntu installer lead? sheesh
[06:11] <mark> yes, clearly :)
[06:12] <cjwatson> yeah, ok, so right before pkgsel would work fine as well
[06:12] <mark> but with many hats things can get intermixed
[06:12] <cjwatson> exactly - Debian installs man-db in base
[06:13] <mark> what does "Build LTSP chroot"?
[06:13] <mark> +do
[06:14] <cjwatson> it's an Edubuntu thing
[06:14] <cjwatson> thin client integration
[06:15] <mark> /dev/sda1                 7.3G    479.8M      6.5G   7% /target
[06:15] <mark> I hope that's gonna be enough :)
[06:15] <cjwatson> if you're installing an LTSP server, it builds a pre-cooked chroot that you can serve out to clients
[06:15] <mark> I see
[06:15] <mark> so I can skip it next time
[06:17] <cjwatson> yeah
[06:17] <mark> ok there we go, I have it log to /target/tmp, and I put a symlink in /var/log/ so it's available over the network
[06:17] <mark> http://yf1014.yaseo.wikimedia.org/stall.trace
[06:17] <mark> gonna start pkgsel now
[06:23] <mark> grr
[06:23] <mark> of course now it went just fine
[06:39] <mark> and again :(
[06:47] <jerom1> bye
[07:11] <mark> can't reproduce it with strace running