Bug#728572: Bug#726123: Side Effect of 726156 Fixes
On Wed, Nov 6, 2013 at 1:07 AM, Michael Vogt <mvo@debian.org> wrote:
> On Sat, Nov 02, 2013 at 12:53:47PM -0500, Martin Gallant wrote:
>> This problem appears to have been exposed by the update to apt
>> 0.9.12.1 (0.9.12.1)
>>
>> I have made three changes to the unattended-upgrade script to dig into this:
>> See enclosed patch.
>>
>> 1. Initialized return code to eliminate referenced before assignment
>> in the exception path
>> 2. Added tracback code to see root cause of error in handler
>>
>> I now get the following traceback information, which points me to 726156:
>
> This looks a lot like Bug#728572 - do you have any hints how to
> reproduce this? Will it be triggered with any package or just with
> specific ones? Its a issue that was supposed to be fixed with apt
> 0.9.12.1. It would be great if you could manually trigger it via:
> $ sudo apt-get install pkgname -o Debug::pkgDPkgProgressReporting=1
> this should give me a good hint what is going on.
>
> Cheers,
> Michael
Just to make sure we are on the same page, my systems are on Jessie, not Sid.
I have both amd64 and i386 Jessie machines, and I see the problem on
both architectures.
$ dpkg -l apt python-apt
ii apt 0.9.12.1 amd64
commandline package manager
ii python-apt 0.9.1 amd64 Python
interface to libapt-pkg
I have never seen any issue when using apt-get.
I only get errors when updating (via cron) through unattended-upgrades.
But note I depend on this automated process, and hardly ever do manual upgrades.
To be clear, the actual installation process always succeeds.
Its the reporting phase that reports errors.
The problem is intermittent in nature.
unattended-upgrades fails about 95% of the time, but occasionally works OK.
My guess is this issue is load dependant, perhaps triggered by large
upgrade jobs.
I have tried downgrading individual packages, and re-upgrading through
unattended-upgrades.
I cannot reproduce the issue when I go trough that path. (Tried 3 packages)
I am still trying to narrow down a single package testcase to reliably
reproduce.
--
Marty
Reply to: