sarnold | juliank: indeed it does look awful; I don't think we've got any ideas either so all advice welcomed | 21:20 |
---|---|---|
juliank | sarnold: Unfortunately no idea :/ | 21:23 |
sarnold | :( | 21:26 |
bdmurray | sarnold: have you heard of lp-bug-dupe-properties? | 21:27 |
sarnold | bdmurray: nope | 21:28 |
bdmurray | sarnold: http://pastebin.ubuntu.com/24911326/ | 21:28 |
sarnold | bdmurray: well that's cool | 21:29 |
bdmurray | I don't know how helpful that specifically is but the tool is neat. | 21:29 |
sarnold | that indeed answers a question I had :) I was hoping there'd be a way to see releases affected and relative weights | 21:29 |
bdmurray | well there you go! | 21:30 |
sarnold | it might be skewed towards new releases because I only manually duped bugs that were filed this month. it might be identical cause with earlier bugs bug something changed "recently" to make this far worse | 21:30 |
sarnold | there's always a trickle of half-installed failed upgrades but this is a real river.. | 21:31 |
bdmurray | It looks like there is a bug with lp-bug-dupe-properties since getting RelatedPackageVersions doesn't work. It might be worth fixing that to find the versions of apt and dpkg. | 21:32 |
bdmurray | This "NULL: ConfigurePending" is weird too. | 21:33 |
infinity | sarnold: The "is not a symbolic link" thing would imply to me that they have a second libssl.so.1.0.0 in their library path, and ldconfig wants to symlink to it, but can't, because the real file exists. | 21:34 |
infinity | sarnold: But why that would suddenly have become an issue is a bit of a mystery. | 21:34 |
sarnold | bdmurray: hrm I think I see loads of NULL: ConfigurePending | 21:34 |
sarnold | it doesn't feel unique here | 21:34 |
sarnold | infinity: oh ewww. | 21:35 |
bdmurray | sarnold: oh you mean across all bugs? | 21:35 |
infinity | sarnold: Though, ldconfig doesn't exit with an error code in that case, so that could be a red herring. | 21:35 |
sarnold | bdmurray: yeah, I know I've seen it in many other bugs | 21:35 |
sarnold | it might still be a contributing factor here, like I said, no ideas.. | 21:36 |
infinity | sarnold: http://paste.ubuntu.com/24911399/ | 21:36 |
infinity | sarnold: I'd posit those postinsts are failing for other reasons that aren't giving you useful terminal output. | 21:36 |
sarnold | :( | 21:37 |
infinity | sarnold: Oh and, indeed, they can't be failing postinsts, because the packages haven't gotten that far. | 21:43 |
infinity | Hrm. | 21:43 |
bdmurray | sarnold: I've confirmed your statement about NULL: ConfigurePending existing in other bug reports | 21:44 |
infinity | sarnold: Plus, libssl-doc and libssl-dev are both represented here. | 21:44 |
juliank | Aren't ldconfig triggers? | 21:44 |
infinity | juliank: Yeah, the ldconfig thing is a red herring here. | 21:44 |
infinity | The real problem is that the packages are half-unpacked. Which is probably why there's a second libssl in the library path. | 21:44 |
sarnold | does this mean dpkg isn't spitting out an error for one of its operations? | 21:45 |
infinity | It could well be a dpkg bug. Are you seeing this on all releases, or...? | 21:45 |
sarnold | infinity: of the ones filed this month, it's 90% 16.04 LTS and 10% 16.10 http://pastebin.ubuntu.com/24911326/ | 21:46 |
infinity | Looks like at least xenial and zesty. | 21:46 |
bdmurray | Preparing to unpack .../libssl1.0.0_1.0.2g-1ubuntu4.8_amd64.deb ... | 21:47 |
bdmurray | Unpacking libssl1.0.0:amd64 (1.0.2g-1ubuntu4.8) over (1.0.2g-1ubuntu4.6) ... | 21:47 |
bdmurray | Log ended: 2017-06-11 18:51:11 | 21:47 |
bdmurray | Should the log have just ended there? | 21:47 |
infinity | Ideally not, but that could be an apt logging bug. :P | 21:47 |
infinity | Which log was that? | 21:48 |
infinity | Unpacking libssl-doc (1.0.2g-1ubuntu4.8) over (1.0.2g-1ubuntu4.6) ... | 21:48 |
infinity | Log ended: 2017-06-08 17:32:24 | 21:48 |
infinity | ^-- From the one that later complains about libssl-doc being half-installed | 21:48 |
bdmurray | https://bugs.launchpad.net/ubuntu/+source/openssl/+bug/1692981/comments/8 | 21:48 |
ubottu | Launchpad bug 1692981 in openssl (Ubuntu) "package libssl-dev:amd64 1.0.2g-1ubuntu11.2 failed to install/upgrade: package is in a very bad inconsistent state; you should reinstall it before attempting configuration" [Undecided,Confirmed] | 21:48 |
infinity | bdmurray: Yup, so that would explain the ldconfig output on that one. | 21:49 |
infinity | bdmurray: If dpkg was interrupted mid-unpack, you'd have libssl.so.1.0.0.dpkg-tmp sitting there. | 21:49 |
bdmurray | Also the other full term log file show libssl-doc being unpacked and ending. | 21:50 |
infinity | And, thus, two versions of the same SONAME. | 21:50 |
infinity | So, in both cases, it looks like dpkg is just dying (or being interrupted by the frontend) | 21:50 |
infinity | Well, s/interrupted/killed/, cause I think a SIGINT would actually unwind gracefully. | 21:51 |
bdmurray | HistoryLog.txt shows "/usr/bin/dpkg exited unexpectedly" | 21:51 |
infinity | Indeed it does. Shame it doesn't log a signal. Though I'm not sure that would be illuminating. | 21:51 |
juliank | Were the upgrades unattended? | 21:52 |
infinity | Commandline: aptdaemon role='role-upgrade-packages' sender=':1.3320' | 21:52 |
infinity | update-manager, maybe? | 21:52 |
infinity | I suppose there's a super tiny chance it could have been OOM killed. But I wouldn't expect that to be extra bad this week compared to last. | 21:53 |
infinity | Unless we recently shipped a memory leak or two. | 21:53 |
* infinity looks at dmesg. | 21:54 | |
bdmurray | All the openssl bugs are amd64 fwiw | 21:54 |
infinity | bdmurray: Could also be a red herring, just because most of our users are amd64. | 21:54 |
bdmurray | right | 21:54 |
infinity | Nothing remotely fun in dmesg. Like, at all. | 21:55 |
infinity | I wonder if they rebooted before apporting. | 21:55 |
sarnold | oh? I thuoght I saw :386 in many of them | 21:55 |
bdmurray | Oh hey - what's this? https://errors.ubuntu.com/problem/fbab11d5ffef5f64e1affa33c6e5bf0330075104 | 21:55 |
slangasek | log on to the internerr | 21:56 |
sarnold | ermagerd internerr! | 21:56 |
bdmurray | sarnold: ack there are some i386 ones | 21:56 |
slangasek | can we kill dpkg-split already | 21:57 |
infinity | I'm blind. Where's the indication that that crash is the same as the bug we were looking at? | 21:58 |
bdmurray | infinity: its not I went looking for dpkg crashes since it died unexpectedly and found this. | 21:58 |
bdmurray | I mean there is no clear indication. | 21:59 |
bdmurray | But if dpkg is crashing on a stable release it wouldn't by default end up in LP. | 21:59 |
infinity | Of course, related or not, that's a whole lot o' crashes. What's the process for getting non-Canonical people access to their crash data? I think guillem should have a poke at that. | 21:59 |
bdmurray | fill out a form somewhere | 21:59 |
bdmurray | https://forms.canonical.com/reports/ | 22:00 |
infinity | Though, that one seems to be fixed in >> xenial? | 22:00 |
infinity | Which means it's probably *not* the bug we're looking at. | 22:00 |
juliank | That error tracker bug looks weird | 22:00 |
bdmurray | only 3 of the reports are using dpkg 1.18.10* | 22:01 |
infinity | That's 3 more than 0. | 22:01 |
infinity | Wait, which reports? | 22:01 |
bdmurray | 3 of the duplicates of the openssl bug in Launchpad | 22:02 |
infinity | For the error tracker, I see everything being <= 1.18.4 | 22:02 |
infinity | The openssl bug, I saw a few in newer releases, yeah. | 22:02 |
infinity | Hence my guess that it's not the same issue. | 22:02 |
juliank | It can't be related because it's a crash that only occurs when formatting an error message when dpkg-split failed. In which case there is nothing to partially unpack | 22:03 |
slangasek | it's not a crash when formatting; it's an abort with message | 22:04 |
slangasek | but still, dpkg-split | 22:05 |
juliank | Ah yes, that makes more sense, it calls abort() there | 22:05 |
infinity | Anyhow, still obviously unrelated, since dpkg-split isn't involved in dpkg --unpack :P | 22:06 |
juliank | Yes, and fixed in newer dpkgs | 22:06 |
juliank | Well, it does not abort() anymore, just logs an error | 22:06 |
bdmurray | Still worth fixing in 16.04 though. | 22:06 |
slangasek | it's worth suppressing those crash reports | 22:07 |
slangasek | it's not really worth fixing; if dpkg-split was invoked at all, your real problem is somewhere upstream | 22:07 |
mapreri | how do you ask autopktest to use multiple triggers? I remember doing so in the past but I seem unable to now | 22:07 |
slangasek | bdmurray: dpkg-split is the tool used to reassemble floppy-disk-sized fragments of a .deb into a single file on your hard drive for installation. It's slightly obsolete | 22:07 |
mapreri | with `&trigger=pkg1/version1,pkg2/version2` it says it's malformed | 22:07 |
infinity | mapreri: trigger=&trigger=&trigger= | 22:08 |
juliank | slangasek: That's dpkg.git commit 521e84da3a2b9ad62d5dbab0f4e1794aef149996 | 22:08 |
slangasek | &mushroom=&mushroom | 22:08 |
mapreri | infinity: ah, thanks | 22:08 |
infinity | SNAKE, AND A SNAKE. | 22:08 |
juliank | https://anonscm.debian.org/cgit/dpkg/dpkg.git/commit/?id=521e84da3a2b9ad62d5dbab0f4e1794aef149996 | 22:08 |
juliank | It was fixed in 1.18.5, so should probably apply easily | 22:08 |
slangasek | bdmurray: so if the best way to suppress the garbage being sent to errors.u.c is by SRUing dpkg, ok - but "fixing" that won't actually benefit users | 22:09 |
infinity | More to the point, we've still got no idea what's eating dpkg on those libssl logs. | 22:10 |
infinity | Which I imagine isn't related to libssl at all. Has anyone found dupes elsewhere? | 22:10 |
bdmurray | infinity: yes | 22:10 |
bdmurray | bug 1692996 | 22:11 |
infinity | Given it's happened to libssl1.0-dev (which is a metapackage), libssl1.0.0 (a library package), and libssl-doc (static docs), it seems like just a timing coincidence. | 22:11 |
ubottu | bug 1692996 in apport (Ubuntu) "package apport 2.20.4-0ubuntu4 failed to install/upgrade: package is in a very bad inconsistent state; you should reinstall it before attempting configuration" [Undecided,Confirmed] https://launchpad.net/bugs/1692996 | 22:11 |
infinity | bdmurray: Okay, that's mildly "comforting". | 22:12 |
infinity | bdmurray: I wonder if the common thread here is that it's all aptdaemon. Though, again, could be a red herring just due to how many aptdaemon upgrades happen. | 22:13 |
infinity | Also, unless we can correlate some actual dpkg *crashes*, it really feel more like something's just murdering it. | 22:14 |
infinity | s/feel/feels/ | 22:14 |
* bdmurray checks his cache o' bugs | 22:14 | |
=== sarnold_ is now known as sarnold | ||
infinity | Fine time for a server to explode. | 22:17 |
infinity | mapreri: What was your full request URI? | 22:18 |
infinity | mapreri: Here's an example of a working on (don't click it :P) | 22:18 |
infinity | mapreri: http://autopkgtest.ubuntu.com/request.cgi?release=artful&arch=amd64&package=snapd&trigger=linux-meta/4.10.0.20.22&trigger=linux/4.10.0-20.22 | 22:18 |
infinity | mapreri: You might just need to urlencode your + and/or ~ in those versions. | 22:19 |
nacc | yeah that's what i've hit in the past (urlencoding) | 22:19 |
infinity | I don't recall the encoding for either of those chars because I haven't done web devel for decades, but I suspect Google can help. :P | 22:19 |
sarnold | mapreri_: <infinity> mapreri: You might just need to urlencode your + and/or ~ in those versions. | 22:19 |
infinity | + = %2B and ~ = %7E | 22:20 |
sarnold | hah excess flood kills too. he probably saw none of this. | 22:21 |
infinity | Oh well. | 22:21 |
* infinity pokes mapreri tentatively. | 22:21 | |
infinity | mapreri: Alive this time? | 22:21 |
* infinity gives up for now until freenode gets over itself. | 22:22 | |
mapreri | http://autopkgtest.ubuntu.com/request.cgi?release=artful&arch=amd64&package=visp&trigger=visp/3.0.1-2build1&trigger=opencv/3.1.0%2Bdfsg1-1%0Aexp1 still says "malformed trigger" :\ | 22:23 |
mapreri | yeah, even if my client says it has quite some lag | 22:23 |
mapreri | I should probably tell my bouncer to try to join channels slower than ~30 at once. | 22:23 |
mapreri | luckily it happens rarely enough ^^ | 22:23 |
juliank | Did nobody write a request-autopkgtest script yet? :D | 22:23 |
sarnold | heh 88 second ping to mapreri :) | 22:23 |
infinity | mapreri: %0A is \n | 22:24 |
infinity | mapreri: Not surprised it doesn't like that. ;) | 22:24 |
infinity | mapreri: ~ = %7E | 22:25 |
bdmurray | infinity: should I be looking for bugs where dpkg exited unexpectedly and aptdaemon is not in use and the package is in an inconsistent state? | 22:25 |
infinity | bdmurray: If it's easier to prove a negative than a positive, yeah. | 22:25 |
infinity | bdmurray: I mean, we want to know one way or the other if *all* related bugs involve aptdaemon, cause then we might have a clue that aptdaemon is murdering its children. | 22:26 |
bdmurray | but if the package is already installed and configured that's not the same issue? | 22:26 |
infinity | bdmurray: But, again, we do so much automagic upgrading with aptdaemon, it could still be a red herring. :/ | 22:26 |
infinity | bdmurray: already-installed-and-configured is a different bug, yes. | 22:26 |
mapreri | Test request submitted. | 22:26 |
mapreri | triggers | 22:26 |
mapreri | ['visp/3.0.1-2build1', 'opencv/3.1.0+dfsg1-1~exp1'] | 22:26 |
mapreri | \o/ | 22:26 |
mapreri | infinity: ♥ | 22:26 |
slangasek | aptdaemon could still be a particular cause of some kinds of failures, e.g. apt debconf pre-configuration | 22:27 |
infinity | bdmurray: This bug would key on half-installed in the terminal log, or dpkg exiting unexpectedly in the dpkg log. | 22:27 |
slangasek | but that would only apply if it were really always libssl1.0.0 getting the axe | 22:27 |
slangasek | ... also libssl1.0.0 has no config script, so nevermind | 22:28 |
infinity | bdmurray: What I really want to know is if we have any dpkg crashes that match the same systems/times as these bugs. If not, then it's more likely a parent committing infanticide. | 22:28 |
infinity | slangasek: It's happened to libssl1.0-dev, which is just a transitional with no maintainer scripts or triggers at all, which I think rules out any packaging issues other than dpkg/frontend being derp. | 22:29 |
* slangasek nods | 22:29 | |
bdmurray | infinity: There's not an easy way to do that but we could ask people to check their whoopsie logs or give us their system identifier. | 22:30 |
infinity | I kinda want to find out that it's actually a memory leak shipped in compiz a week earlier and dpkg is being OOM-killed, just because that would be almost as much fun as Can't Print on Tuesdays. | 22:32 |
* infinity tries to find one with an intact dmesg that wasn't obviously rebooted before the report was sent. | 22:34 | |
infinity | The fact that all of them (so far I've been through about 10) seem to have been rebooted before the report was sent. Is that coincidence due to how/when apport runs, or could that be indicative of a larger issue that forced a reboot? | 22:35 |
infinity | bdmurray: ^ | 22:35 |
infinity | Cause another suspicion here is that weird "cannot fork" thing I saw on my machine a while back. | 22:36 |
infinity | Which would, indeed, lead to dpkg bailing during unpack. | 22:36 |
sarnold | infinity: you ran into a 'cannot fork' in the wild? as root? | 22:37 |
infinity | sarnold: Hard to say if I ran into as root, given that I couldn't execute anything to elevate privs. | 22:38 |
infinity | sarnold: So, "maybe"? :P | 22:38 |
sarnold | infinity: hehe | 22:40 |
sarnold | infinity: I thought maybe that was apt/dpkg ... | 22:40 |
infinity | Anyhow, every one of those dupes has a fresh dmesg. Which might mean something, or might just mean apport doesn't do useful things until reboot. | 22:42 |
infinity | In which case, sending dmesg instead of kern.log.* seems mostly useless. | 22:42 |
sarnold | I'm not sure about the relative merits of kern.log vs dmesg but I close a ton of bugs are hardware related thanks to dmesg errors | 22:43 |
bdmurray | An interesting thing about apport and whoopsie bug reports is that they add what's in /var/crash and I'm not seeing any dpkg crashes there. | 22:47 |
infinity | bdmurray: Which might well imply that dpkg is being killed, not crashed. | 22:49 |
sarnold | Bug #1699356 is libgcc1:i386 -- the dpkg history log shows "Sub-process /usr/bin/dpkg returned an error code (1) | 22:50 |
ubottu | bug 1699356 in gcc-6 (Ubuntu) "package libgcc1:i386 1:6.2.0-5ubuntu12 failed to install/upgrade: package libgcc1:i386 is not ready for configuration cannot configure (current status 'half-installed')" [Undecided,New] https://launchpad.net/bugs/1699356 | 22:50 |
sarnold | .. before the "Sub-process /usr/bin/dpkg exited unexpectedly" | 22:51 |
mapreri | could anybody merge doxygen? doxygen-latex is uninstallable in proposed due to latex being rude at the world, and uninstallable doxygen-latex makes me sad. | 22:53 |
mapreri | (happy to provide debdiffs and all, but `grab-merge doxygen` gets it right by itself…) | 22:53 |
mapreri | actually, grab-merge is not enough, as the latest upload is not yet there | 22:56 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!