CIA-7 | debian-installer: adconrad * r1715 ubuntu/ (build/config/armel/omap4.cfg debian/changelog): Move omap4 kernels to 3.4.0-203. | 03:28 |
---|---|---|
jibel | xnox, cjwatson if there is no fix for bug 1020574, is there a possibility to revert apt and python-apt (if the problem comes from here) to make the images working again ? | 08:06 |
ubot2 | Launchpad bug 1020574 in ubiquity "SystemError: Broken pipe while installing language packs" [Critical,Confirmed] https://launchpad.net/bugs/1020574 | 08:06 |
cjwatson | haha, good luck with that | 08:25 |
cjwatson | I don't think that's terribly plausible TBH | 08:25 |
xnox | cjwatson: well i *think* i can comment the offending line out of apt (it's just a status update message) to unbreak ubiquity.... but it's still weird why that pipe is closed, it is used to communicate progress back to ubiquity | 08:27 |
xnox | 17339 15:57:48.385417 write(79, "pmstatus:dpkg-exec:0:Running dpk"..., 34 <unfinished ...> | 08:27 |
xnox | 20429 15:57:48.385458 <... read resumed> "pmstatus:dpkg-exec:0:Running dpk"..., 4096) = 34 <0.143676> | 08:27 |
xnox | 17339 15:57:53.068803 write(79, "pmstatus:firefox-locale-en:2:Pre"..., 57) = -1 EPIPE (Broken pipe) <0.000025> | 08:27 |
cjwatson | is there a useful subset of the strace you can post somewhere? | 08:28 |
cjwatson | I would be happy to help analyse it | 08:28 |
cjwatson | 900M is a bit much, but maybe something that starts from the common ancestor of pids 17339 and 20429 | 08:30 |
xnox | it's only 60MB when xz compressed ;-) | 08:31 |
xnox | http://paste.ubuntu.com/1077753/ | 08:31 |
xnox | i should have split the strace per pid/fork | 08:31 |
xnox | ok, let me try to grep that subset. | 08:32 |
brendand | cjwatson, any luck reproducing bug 1021293? | 08:33 |
ubot2 | Launchpad bug 1021293 in checkbox-satellite "Ubuntu 12.04 install stalls when doing apt-get upgrade" [Undecided,New] https://launchpad.net/bugs/1021293 | 08:33 |
cjwatson | brendand: not as yet sorry | 08:33 |
cjwatson | xnox: oh, 60M is practical for me to download if it's practical for you to upload | 08:34 |
cjwatson | better to have the whole thing if possible | 08:34 |
xnox | cjwatson: link in the bug 1020574 | 08:34 |
ubot2 | Launchpad bug 1020574 in ubiquity "SystemError: Broken pipe while installing language packs" [Critical,Confirmed] https://launchpad.net/bugs/1020574 | 08:34 |
xnox | cjwatson: you can even ssh into the box.... | 08:34 |
xnox | http://people.canonical.com/~xnox/ubiquity.trace.xz | 08:35 |
cjwatson | ETA 4min | 08:35 |
xnox | cjwatson: of the fix uploaded? WOW you are quick ;-) | 08:36 |
* xnox kidding | 08:36 | |
cjwatson | hah, no | 08:37 |
cjwatson | well, OK, so the process reading from that pipe has exited | 08:52 |
cjwatson | it never does very much; I wonder what it is | 08:53 |
xnox | - write(OutStatusFd, status.str().c_str(), status.str().size()); | 08:54 |
xnox | + FileFd::Write(OutStatusFd, status.str().c_str(), status.str().size()); | 08:54 |
xnox | this is the code in apt, which prints the status line back. | 08:54 |
xnox | ... it was changed.... | 08:54 |
cjwatson | Yeah, but my suspicion currently rests on the other end of the pipe | 08:55 |
xnox | ok. | 08:55 |
cjwatson | This sort of thing can arise from a missing close-on-exec flag | 08:55 |
cjwatson | But just a guess so far | 08:55 |
* cjwatson has sudden nostalgia for his first real job interview, which involved on-the-fly strace analysis | 08:56 | |
cjwatson | 17339 is plugininstall.py | 08:56 |
cjwatson | string lengths in this strace are unfortunately short, but I guess that was for practicality? | 08:57 |
cjwatson | makes it hard to find some things | 08:57 |
xnox | well, it took ~2-3h to get the stack trace. cause ubiquity was slow under it. I can run it again if you know options which will reduce the cruft and increase additional info. | 09:01 |
xnox | sorry, if I missed stuff.... | 09:01 |
cjwatson | I usually bump -s to something bigger, maybe 512 for this; but don't worry for now | 09:02 |
cjwatson | So, we're in the call to cache.install_archives in commit_with_verify, clearly | 09:03 |
cjwatson | The PROGRESS START message is therefore from DebconfInstallProgress.startUpdate | 09:03 |
cjwatson | So the clone of 20429 is from something called immediately after the startUpdate hook | 09:04 |
cjwatson | Ah, which is .run of course | 09:04 |
cjwatson | Hm, it strikes me as a little odd that we're using "while self.update_interface():" but update_interface does not generally seem to return a vaue | 09:08 |
cjwatson | *value | 09:08 |
* cjwatson looks at older python-apt | 09:08 | |
cjwatson | So that might well be the real source of the problem; but how are you supposed to tell from update_interface when it's done? | 09:13 |
xnox | def update_interface(self): | 09:13 |
xnox | apt.progress.base.InstallProgress.update_interface(self) | 09:13 |
xnox | self.frontend.refresh() | 09:13 |
xnox | wrapped in ubiquity... | 09:14 |
cjwatson | Also wouldn't hurt to update the API | 09:15 |
cjwatson | xnox: Not relevantly, that's in auto_update.py | 09:15 |
xnox | cjwatson: yeah... | 09:16 |
cjwatson | And even then, I'm having trouble seeing a way to detect EOF without copying all of update_interface, not just wrapping it | 09:16 |
CIA-7 | ubiquity: cjwatson * r5534 trunk/ (debian/changelog ubiquity/install_misc.py): Update DebconfInstallProgress method names for the python-apt 0.8 API. | 09:17 |
cjwatson | Ah - python-apt itself uses a select loop | 09:18 |
cjwatson | Not sure why it bothers with a timeout | 09:21 |
cjwatson | Maybe the simplest answer is (a) use a select loop rather than spinning (b) don't worry about a sensible exit condition and instead just kill the status-to-debconf subprocess when pm.do_install completes | 09:27 |
cjwatson | xnox: You're correct, I think, that the apt change you pointed to above exposed this problem, but it was never correct in ubiquity AFAICS | 09:28 |
cjwatson | And this probably resulted in horrendous inefficiency and/or wildly incorrect progress reporting | 09:28 |
cjwatson | So reverting apt is the wrong answer; we should instead fix the bug it exposed | 09:29 |
cjwatson | I'm thinking http://paste.ubuntu.com/1077814/ | 09:30 |
xnox | hmm... do you want me to test it? | 09:32 |
cjwatson | If you can, though I'm trying too | 09:33 |
cjwatson | Also a sanity check :-) | 09:33 |
cjwatson | (I need to confirm that I can reproduce the bug first) | 09:33 |
xnox | =0))) | 09:34 |
cjwatson | Hmm, that patch gets me to the slideshow with no progress messages | 10:13 |
xnox | cjwatson: you did add missing imports right?! =) | 10:14 |
cjwatson | plugininstall isn't even running ... | 10:14 |
xnox | import select; import signal... | 10:14 |
cjwatson | xnox: err, *cough*, I was misled by pyflakes.vim not telling me about them :) | 10:14 |
cjwatson | thanks | 10:14 |
xnox | cjwatson: i'm off to the office. we'll be back online in ~40minutes | 10:16 |
cjwatson | right. I have to wait for dpkg to be fixed anyway before we can build anything | 10:17 |
cjwatson | bah, syntaxerror too | 10:19 |
cjwatson | better now | 10:47 |
CIA-7 | ubiquity: cjwatson * r5535 trunk/ (debian/changelog ubiquity/install_misc.py): | 10:48 |
CIA-7 | ubiquity: Rearrange DebconfInstallProgress to not assume that update_interface | 10:48 |
CIA-7 | ubiquity: returns a value (LP: #1020574). | 10:48 |
CIA-7 | ubiquity: cjwatson * r5536 trunk/ (d-i/manifest debian/changelog): | 10:51 |
CIA-7 | ubiquity: Automatic update of included source packages: flash-kernel | 10:51 |
CIA-7 | ubiquity: 3.0~rc.4ubuntu5, partman-base 158ubuntu1, partman-target 78ubuntu1, | 10:51 |
CIA-7 | ubiquity: preseed 1.53ubuntu1. | 10:51 |
CIA-7 | ubiquity: cjwatson * r5537 trunk/debian/changelog: releasing version 2.11.9 | 10:58 |
xnox | cjwatson: ubiquity became far more chatty now. "removing this" "configuring that" | 12:26 |
cjwatson | yeah, it would do | 12:27 |
xnox | previously it seemed to me to stick at "installing..." during the langpack installation step =) | 12:27 |
xnox | i like it, previously I had to expand the thinggy to see what is going on =) | 12:27 |
xnox | cjwatson: is the twitter feed not running intentionally? Can we have it on hashtags #quantal and #12.10 ? | 12:28 |
xnox | too user friendly, geeks need scrolling text ;-) | 12:28 |
cjwatson | no intention on my part | 12:29 |
cjwatson | feel free to make sensible adjustments :) | 12:29 |
xnox | cjwatson: well LVM auto-install now succeeds. Upon reboot it drops into initrd shell =) I bet the lvm2 modules were not added... and initramfs was not regenerated.... | 12:31 |
cjwatson | IIRC d-i does that in base-installer, so you'd need to duplicate that | 12:36 |
cjwatson | maybe via register-module? I don't quite remember | 12:37 |
cjwatson | hmm. looking now I can't find where it's done. | 12:38 |
cjwatson | the lvm2 initramfs hook, I guess. so that's odd because I thought we unconditionally regenerated the initramfs. | 12:39 |
cjwatson | maybe an ordering thing? | 12:39 |
cjwatson | is the lvm2 package installed in the target? | 12:40 |
cjwatson | ah - perhaps you need to copy partman-lvm's post-base-installer hook somewhere useful for ubiquity, or duplicate it | 12:41 |
cjwatson | this is a bit tedious, there's piles of crap in debian/rules that deals with similar kinds of stuff, or sometimes it's duplicated in components or scripts | 12:41 |
xnox | cjwatson: right. I think it's because lvm2 was not in the squashfs and instead it's just "installed" in the live session. | 13:07 |
xnox | because the update-initramfs is done at the end of the install. | 13:07 |
xnox | and lvm binaries did not appear in the list of "things not copied into the target" | 13:08 |
xnox | and it was not installed on the target | 13:08 |
xnox | cjwatson: shall I remaster the ISO or should I do a merge proposal with my work? | 13:09 |
xnox | cjwatson: after chrooting into mounted target /boot and / and doing # apt-get install lvm2 & doing a reboot it works like a charm =) | 13:09 |
xnox | cjwatson: cause I don't see anything special done, for e.g. btrfs-tools | 13:10 |
* xnox maybe I should remaster an ISO for training purposes. | 13:11 | |
cjwatson | xnox: I'd like to see an MP | 13:16 |
xnox | cjwatson: ok. | 13:16 |
cjwatson | is apt-install called for lvm2? | 13:16 |
cjwatson | you can tell by looking in /var/lib/ubiquity/apt-installed | 13:16 |
xnox | cjwatson: in the target or the live cd during installation? | 13:26 |
xnox | (i finished the install and have /var/log/installer/*) | 13:26 |
xnox | nothing in after logs =( | 13:29 |
xnox | only /var/lib/ubiquity/os-prober-cache left. | 13:29 |
cjwatson | in the live cd during installation | 13:32 |
cjwatson | well, at the end, before rebooting | 13:32 |
xnox | ok. let me double check. | 13:34 |
xnox | cjwatson: mpt has finished: manual & automatic crypt; manual lvm. | 13:35 |
xnox | so that's what I am aiming for the feature freeze. | 13:36 |
xnox | cjwatson: will that be enough to drop alternative? (no raid...) | 13:36 |
cjwatson | hmm, not sure | 13:40 |
cjwatson | maybe, since desktop RAID (not DM-RAID) isn't a big thing really | 13:40 |
brendand | cjwatson - we have lab presence again. is there any extra info you'd like regarding the upgrade bug? | 13:40 |
cjwatson | might be worth looking around to see if people commonly report using the alternate CD in order to use RAID | 13:41 |
cjwatson | brendand: not right now, thanks. sorry, my day has been (a) critical ubiquity bug in quantal (b) server CD -> squashfs work requested increasingly urgently by Daviey (c) working on a regression in recent Launchpad changes (d) this upgrade bug | 13:41 |
brendand | cjwatson, i understand | 13:42 |
gema | smoke testing bug, easy to fix I hope, bug 1021718 | 13:45 |
ppisati | hi guys | 13:45 |
ubot2 | Launchpad bug 1021718 in debian-installer "Ubuntu Precise ISO test failed in Jenkins due to debian installer failed to get debconf answer 'base-installer/kernel/linux/initrd'." [Undecided,Confirmed] https://launchpad.net/bugs/1021718 | 13:45 |
ppisati | http://people.canonical.com/~ppisati/Jul062012_omap3_preinstall_brk/ | 13:45 |
ppisati | while trying preinstall server omap3 | 13:45 |
ppisati | ubiquity.install_misc.InstallStepError: Plugin console_setup failed with code 1 | 13:45 |
ppisati | i've both syslog and oem-config.log | 13:45 |
ppisati | the installer kept restarting | 13:46 |
mpt | cjwatson, next week I'll polish off automatic crypt design and then go on to RAID design. | 13:46 |
ppisati | until it gave up and landed me in a # console | 13:46 |
xnox | ogra_: i think you might want to see ppisati question ^^^^ | 13:46 |
ogra_ | xnox, i sent him here :) | 13:46 |
cjwatson | gema: stuck on a soyuz bug / new kernel upload | 13:47 |
cjwatson | not an installer bug | 13:47 |
gema | cjwatson: who should I notify? | 13:47 |
cjwatson | the kernel team already know about it | 13:47 |
gema | cjwatson: ack | 13:47 |
cjwatson | I'll nag them | 13:47 |
gema | thanks | 13:47 |
xnox | ogra_: ping pong eh? =) | 13:48 |
xnox | ppisati: can you pastebin the logs? | 13:48 |
ogra_ | xnox, well, we were in #ubuntu-arm but i think its not arch specific at all | 13:48 |
ppisati | xnox: available here - http://people.canonical.com/~ppisati/Jul062012_omap3_preinstall_brk/ | 13:48 |
cjwatson | how's it not arch-specific? | 13:48 |
cjwatson | Jun 27 02:39:55 localhost ubiquity: Unsupported platform. | 13:48 |
cjwatson | Jun 27 02:39:55 localhost ubiquity: run-parts: /etc/initramfs/post-update.d//flash-kernel exited with return code 1 | 13:48 |
xnox | ppisati: ideally run ubuntu-bug ubiquity. or paste. Oh ok =) | 13:49 |
ogra_ | console_setup exploding is arch specific ? | 13:49 |
ogra_ | (unsupported arch is fine, just fixed that (pending upload)) | 13:49 |
cjwatson | console-setup calls update-initramfs | 13:49 |
cjwatson | update-initramfs explodes because the flash-kernel hook explodes | 13:49 |
ogra_ | oh, sigh ... yeah | 13:49 |
ogra_ | sorry | 13:49 |
cjwatson | the top-level error message is just a bit misleading, that's all | 13:49 |
ogra_ | yep, should have thought about the initramfs bits | 13:50 |
ppisati | cjwatson: there some more errors in oem-config | 13:50 |
ppisati | sh: 1: killall.sh: not found | 13:50 |
ppisati | sh: 1: rdisc6: not found | 13:50 |
ppisati | etcetc | 13:50 |
ppisati | is it ok? | 13:50 |
cjwatson | ppisati: don't care | 13:50 |
cjwatson | gema: actually, from https://launchpad.net/~canonical-kernel-team/+archive/ppa/+packages I see that it's already building, so no need/point nagging the kernel tam | 13:51 |
ppisati | ack | 13:51 |
cjwatson | *team | 13:51 |
gema | cjwatson: awesome, thanks | 13:51 |
cjwatson | ppisati: that's just bits of netcfg that (a) don't work outside d-i but (b) don't really matter here either | 13:51 |
ppisati | ok | 13:52 |
cjwatson | gema: https://irclogs.canonical.com/2012/07/06/%23launchpad-ops.html#t07:21 has the gory details if you care :) | 13:54 |
cjwatson | (hideous LP internals) | 13:54 |
gema | cjwatson: thanks! | 13:55 |
* cjwatson grumbles and queue-jumps those PPA builds | 13:58 | |
xnox | cjwatson: so lvm2 is not in /var/lib/ubiquity/* but in the apt-removed there is cryptsetup. | 14:07 |
* xnox mouse is stuck =( | 14:09 | |
cjwatson | ok, that sounds like you need to hunt down why apt-install wasn't called then; I thought partman-lvm or some such did that | 14:12 |
xnox | ok. | 14:13 |
* ogra_ sends one of his cats to xnox (that will teach the mouse !) | 14:13 | |
xnox | cjwatson: well apt-install was called, with a few packages, but not lvm2. but yeah, checking now. | 14:14 |
xnox | I am not going to ask why livecd has /lib/systemd/ | 15:20 |
* xnox *sigh* | 15:20 | |
cjwatson | bunch of packages shipping systemd service files as well | 15:21 |
xnox | cjwatson: so, partman-btrfs does apt-install in the partman/finish.d script | 15:28 |
xnox | but crypto,lvm,raid do it in the scripts shipped under /usr/lib/post-base-installer.d/ | 15:28 |
xnox | cjwatson: should I add a plugin to iterate across scripts in the /usr/lib/post-base-installer.d? | 15:28 |
xnox | or should I shove those scripts into partman/finish.d for ubiquity to pick up? | 15:29 |
xnox | or is btrfs actually apt-installed from partman/finish.d in ubiquity? | 15:29 |
* xnox looks what you said above | 15:29 | |
cjwatson | standard (perhaps a bit shonky) practice is to do something like invoking them from an InstallPlugin (or whatever it's called) | 15:30 |
xnox | let me look. | 15:30 |
cjwatson | I was a bit reluctant to iterate across everything since (a) that path is nasty for ubiquity and ought to be under /usr/lib/ubiquity/ or something (b) I wasn't sure all the p-b-i scripts were sane (c) you'd still have to have debian/rules or similar code to arrange for them to be installed | 15:31 |
cjwatson | partman/finish.d should be run automatically | 15:31 |
cjwatson | there is an argument that at least the lvm one belongs in finish.d | 15:32 |
cjwatson | I haven't looked at the others | 15:32 |
cjwatson | certainly moving to finish.d would be least code | 15:32 |
xnox | what is finish-install.d vs finish.d vs post-base-installer.d ? where can I check when what is called? | 15:34 |
cjwatson | finish.d is called from partman itself (well, partman-commit in ubiquity) | 15:44 |
cjwatson | post-base-installer.d and finish-install.d are both d-i-specific; they're called, respectively, at the end of base-installer and at the very end of installation | 15:45 |
xnox | partman-crypto uses all three: finish.d post-base-installer.d and finish-install.d | 15:45 |
xnox | and e.g. d-i-utils has post-base-installer.d to add/blacklist modprobe modules | 15:46 |
cjwatson | there's a d-i internals document linked from Installer/Development that IIRC goes through all this | 15:51 |
xnox | thanks, let me find it. | 15:54 |
xnox | cjwatson: the p-b-i does ugly crap, like raising pointless debconf questions and confirmation dialogs | 15:54 |
Sargun_Screen | How would one update a package on the netinstall (PXE) - specifically if they were running it as a netboot | 20:53 |
Sargun_Screen | The mdadm package causes my kernel to fail | 20:54 |
cm-t | omg! I just successfully installed an array list of package (vlc, vrms) with ubiquity, I was so making something far while the code was allready there ! | 21:07 |
cm-t | (hi) | 21:07 |
stgraber | apt-install vlc vrms | 21:07 |
stgraber | :) | 21:07 |
cm-t | i mean | 21:07 |
cm-t | getting the good /target | 21:07 |
cm-t | it was just self.do_install(packageArray) | 21:08 |
cm-t | >.< | 21:08 |
cm-t | was trying so many thing, and did not even scroll at the bottom of the script where this wonderfull method was | 21:09 |
cm-t | just need to work this week-end with my group mates to get this with oneconf (they made I think, just some sso problems) and we might consider having something running (not talking about a clean code for the while ^^) | 21:11 |
cm-t | I will update blueprint this week-end cjwatson, mpt | 21:12 |
infinity | cjwatson: Say, opinion. Since we fixed #1010708 in -proposed (and soon, updates), and since those files/directories are managed in a dubious and poorly-tracked manner anyway, do you have any objections to just "fixing" the bug in the release pocket by moving the two uI* files where they belong, so there isn't inconsistency between release and updates (and later releases)? | 21:23 |
infinity | cjwatson: Oh, I guess that would break the manifest too. Hrm. | 21:24 |
infinity | cjwatson: Oh, wait. But the *SUMS could just be fixed and re-signed. That doesn't seem like an awful solution for a port in flux. Worth a ponder. | 21:25 |
cjwatson | I'm not at all wild about changing the release pocket of a stable release, although I guess the number of people who would notice or care would be small | 22:02 |
cjwatson | the *SUMS files would be automatically re-signed if they change | 22:02 |
infinity | Yeah, I figured that. | 22:24 |
infinity | And I'm not a big fan of changing release pockets either, but this isn't a Packages file or anything, and it seems like minimal impact to avoid confusion. | 22:24 |
infinity | I think I'll JFDI and ask for forgiveness. :P | 22:24 |
cjwatson | Yeah. | 22:31 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!