aptdaemon crashes during update-manager run: "Could not cancel transaction"

[Edward Donovan]

This crash is the subject of bug 628104 and bug 855394. Those are marked fixed but it's still happening. The latter bug was closed, with this comment:

  "As we have not received any duplicates of this bug since January and the bug pattern has received no hits, I am going to mark this bug report as Fix Released for Precise and set the Oneiric task to Won't Fix. In the event that anyone encounters this bug in the future please report a new crash report."

This is that bug report. :)

Brian also wrote:

  "There really was no clear indication, a changelog for example, that bug 628104 was really fixed."

That might be true.

A number of users have tried to report the crash through apport, but the traceback has always been duped to bug 628104. Which none of us can re-open.

I myself cannot reproduce it. The recent information is all at 628104. Thanks.

[Steve Langasek (vorlon)]

Edward,

The request was that people file new *crash* reports; i.e., to use the apport interface to report a new bug which would show us the crash. You write:

> I myself cannot reproduce it.

Which makes this bug report of dubious value.

> A number of users have tried to report the crash through apport, but the traceback has always been duped to bug 628104.

The last such duplicate was filed in May 2011. There have been no duplicate bugs reported since then.

In your first comment on bug #628104, you say:

> Apport just sent me here, from an apparent crash,

But how exactly did apport send you there? There is no bug pattern in place for bug #628104 and never has been. Nor did you file a new bug report and have it marked as a duplicate by the retracer. Did apport really send you there at all, or did you simply follow a link to that bug in your browser when offered as a "possibly similar" existing bug?

The only new bug report filed against update-manager or aptdaemon showing this crash is bug #976889, which was filed a week ago.

So I don't see any evidence that bug #628104 should be reopened, and this new bug report doesn't contain any actionable information.

[Edward Donovan (edward.donovan)]

Hi Steve, thanks very much for responding.

Your questions demonstrate that I do not yet know how apport and LP match duplicates, and more, and I may be using some terms poorly.

> But how exactly did apport send you there?

Apport popped up its window and asked me to file. I clicked 'Continue' on the local apport client, and it told me the bug was already reported, and opened bug 628104. I think the client-side duplicate checking matched it.

> There is no bug pattern in place for bug #628104 and never has been.

I see I don't know the bug system well enough. It looks like it's matching the traceback, with the several lines of python error output, ending in "AptDaemonError: org.debian.apt: Could not cancel transaction". I don't yet know the difference between that and a proper bug pattern, then.

> Nor did you file a new bug report and have it marked as a duplicate by the retracer.

I did not, since I saw other users had done so, and the retracer had duped them back to 628104. I intended to file that bug, then I realized it had been done, as in comments 7 and 8, there:

  "When the crach occured the system claimed it was already reported and opened this bug.

  I now manually uploaded it as Bug #915413"

  ------

  "I had this crash occur in precise as well. Note that #915413 was marked as a duplicate of this bug."

I tried to follow what the system wanted, and uploaded my crash file there. (https://launchpadlibrarian.net/99074345/_usr_bin_update-manager.1000.crash)

Eventually, it seemed I had to try filing this bug, purposefully without the traceback, to make an open bug report.

> Did apport really send you there at all, or did you simply follow a link to that bug in your browser when offered as a "possibly similar" existing bug?

Yes, apport sent me there directly, same as the user I quote above. I've seen the "possibly similar" screen, but apport did not have these questions in this case. The recent comments on 628104 look like they're generally coming straight via apport.

> The only new bug report filed against update-manager or aptdaemon showing this crash is bug #976889, which was filed a week ago.

Ah, I can't see that one; don't know if it's private, or what.

Rather than new reports, there are a number of users filing new comments on 628104, because apport and/or LP send them there. That's what I'm being dogged about here -- it appears users now can't file a new bug, if they follow apport. Nor open the old one So I've been trying to get human attention, in case it's needed.

It has appeared, and it's only my best effort to understand, that new reports are falling into a little hole in the system. I'm rather sorry if I seem to come off as a crank. I've seen the new reports slowly accumulating on that bug, and wondered if it was my little user duty to do more. Thanks.

[Brian Murray (brian-murray)]

There was a bug pattern for bug 855394 though which is similar.

[Steve Langasek (vorlon)]

Brian,

> There was a bug pattern for bug 855394 though which is similar.

Isn't that the one that we saw no hits against via bit.ly?

[Brian Murray (brian-murray)]

Yes, that's true and if Edward would have hit the pattern then the count would have been incremented by one.

[Edward Donovan (edward.donovan)]

Hi gentlemen -

I still do not grok bug patterns enough; excuse me. For me and other recent reporters, it's the traceback which matches bug 628104, and is automatically duped to it. The client-side duping catches it; if you manually upload the crash, the server-side retracing catches it.

You can see a number of reports there, in recent months. Ignore my multiple comments, since I'm summarizing here, and since most are outdated now, but I can't delete them.

For some, like me, the problem was transient and unimportant. Others seem to have their upgrades broken by it, but they have not reported back, nor subscribed.

If we cannot find any lasting breakage when this hits, it may be a matter of an uncaught exception, that we might handle more routinely.

[Edward Donovan (edward.donovan)]

I can now reproduce this, and might rewrite the description, and even reopen the status. But I will wait a while to hear back in my last comment, before trying that.

In brief, to reproduce it, quit update-manager from the Launcher quicklist, during a running upgrade. It may not happen 100% of attempts, but should only take a couple tries before starting apport, and leading to 628104, as described above. Thank you.

[Edward Donovan (edward.donovan)]

I have read up a little on bug patterns. Since apport and LP will match on tracebacks, and did so, here, that still seems significant, though there is no pattern written. (As I understand it.) No?

Thanks.

[Steve Langasek (vorlon)]

On Sat, Apr 21, 2012 at 01:25:29AM -0000, Edward Donovan wrote:
> I still do not grok bug patterns enough; excuse me. For me and other
> recent reporters, it's the traceback which matches bug 628104, and is
> automatically duped to it. The client-side duping catches it; if you
> manually upload the crash, the server-side retracing catches it.

The issue is that we have no idea what you're referring to here as
"client-side duping". We have bug patterns, which tell apport not to file a
bug report but direct the user to a web page instead; but there is no bug
pattern in place for this bug. There's the apport crash retracer which
identifies duplicates, but that only happens *after* a new bug report has
been filed, and we see no record of new bug reports being filed.

So we frankly have no idea what you're seeing, because it doesn't match the
developers' understanding of how apport works. I think you will need to
walk us through, page-by-page and quoting word for word, what you see on
your system when the crash report appears, so we can understand exactly
what's going on.

> You can see a number of reports there, in recent months. Ignore my
> multiple comments, since I'm summarizing here, and since most are
> outdated now, but I can't delete them.

Bug #628104 has new comments, but it has never had a bug pattern and there
are no duplicate bugs against it since May 2011. Bug #855394 had a bug
pattern which was being used to track hits, and received none from January
to the end of March. So while I believe that apport directed you here, I
don't know *how*, and we need to fix that so that we can get good, fresh
crash reports from the users who are seeing the issue.

[Edward Donovan (edward.donovan)]

Thank you, Steve, that helps quite a bit.

The 'client-side duping' I'm talking about is, I guess, a fairly recent arrival?

  http://www.piware.de/2011/11/apport-1-90-client-side-duplicate-checking/

"So with the just released Apport 1.90 we introduce client-side duplicate checking. So from now, when you report a crash, you are likely to see “We already know about this” right away, without having to upload or type anything, and you will get directed to the bug page. You should mark yourself as affected and/or subscribe to the bug, both to get a notification when it gets fixed, and also to properly raise the “hotness” of the bug to bubble up to developer attention."

Does this help explain my preceding posts? If not, excuse me again! :)

To me, that's how I got to 628104, and from the looks of it, I think that's true for just about all the newer comments, that reference Precise. From what I can see, people are mostly not subscribing, as Martin proposes. That may be creating, inadvertently, some hit-and-run reporting, with not much follow-up. (It's this impression that has led me to be sort of knuckleheadedly persistent, in my fumbly efforts to pursue this.)

I'll try to post more tomorrow, with more info about reproducing, and I'll see if this has made any better sense of things, yet. Thanks.

[Edward Donovan (edward.donovan)]

To be more explicit, I think the recent reports like mine, against 628104, *are* the new duplicates -- their equivalents, since the system intercepts them. (It's been my plan to mail Martin, to ask about this sort of thing, after release happens.)

Since we can't open the bug we're sent to, bug 915413 was a user's subsequent attempt to make an open bug, manually. It got re-traced and duped to our other friend, bug 855394. So this bug was my attempt to make an open bug, and I knew I had to omit the traceback. (See, it's all so simple. Wait, it's not. ;)

For reproducing, I guess my comments in #7 cover it, but I can write that out in more of a list, if that makes it easier to read.

Since I have the good luck of getting a slice of your attention, Steve, let me ask you a question about releasing, if you will. I'm chasing a few different bugs right now. I don't know whether this is a time to work urgently, for a remote chance of making Precise. Or whether that has really passed, for anything that isn't truly massive. In that case, if anything, I would hold back for this week, and leave the developers some quiet. And personally calm down a bit.

Thanks again.

[Steve Langasek (vorlon)]

On Sun, Apr 22, 2012 at 03:54:13PM -0000, Edward Donovan wrote:
> Steve, let me ask you a question about releasing, if you will. I'm
> chasing a few different bugs right now. I don't know whether this is a
> time to work urgently, for a remote chance of making Precise.

Any fixes in this area will not make it for 12.04.0. They will be
considered for inclusion in 12.04.1.

[Steve Langasek (vorlon)]

On Sat, Apr 21, 2012 at 07:46:37AM -0000, Edward Donovan wrote:
> The 'client-side duping' I'm talking about is, I guess, a fairly recent
> arrival?

> http://www.piware.de/2011/11/apport-1-90-client-side-duplicate-
> checking/

> "So with the just released Apport 1.90 we introduce client-side
> duplicate checking. So from now, when you report a crash, you are likely
> to see “We already know about this” right away, without having to upload
> or type anything, and you will get directed to the bug page. You should
> mark yourself as affected and/or subscribe to the bug, both to get a
> notification when it gets fixed, and also to properly raise the
> “hotness” of the bug to bubble up to developer attention."

> Does this help explain my preceding posts? If not, excuse me again! :)

Yes it does, thanks very much. I've asked Martin for input regarding the
client-side duping being untraceable by developers.