Local testing of imports

[Diogo Matsubara]

On Fri, Dec 11, 2009 at 1:39 AM, James Westby
<james.westby at canonical.com> wrote:
> On Fri Dec 11 02:06:11 +0000 2009 Jonathan Lange wrote:
>> > In addition, I think it would be useful if we could classify the
>> > list of current failures. Does the LP OOPS system have code to
>> > match tracebacks? Anyone know of anything else to do this?
>>
>> I don't think so, but then I'm not really sure what you mean.

I don't understand what you mean.
Let me try to explain what oops-tools[1] does and then you'll be able
to tell me if it could be used for your use case.

All oops generated by any of the LP application servers are processed
daily and summarized on what's called an OOPS summary. You can see an
example of one those summaries in:
https://devpad.canonical.com/~lpqateam/oops-summaries/lpnet-2009-11-30.html

Look at the All exceptions section:

 55 AttributeError: 'NoneType' object has no attribute 'email'
    Bug: https://launchpad.net/bugs/452525
    POST: 45 GET: 10  Robots: 0  Local: 34
      19 https://launchpad.net/branchfilesystem/translatePath
(BranchFileSystemApplication:BranchFileSystem)
        OOPS-1430XMLP13, OOPS-1430XMLP14, OOPS-1430XMLP15,
OOPS-1430XMLP16, OOPS-1430XMLP17

oops-tools then aggregates all OOPSes that have the same exception
type (e.g. AttributeError) and exception value (e.g. 'NoneType' object
has no attribute 'email'), displaying a sample of those OOPSes along
with the URL where that was triggered (and other information as you
can see above).

Once an specific OOPS instance (e.g.OOPS-1430XMLP13) is analysed by
one of the LP engineers, a bug is filed and then linked to the OOPS
report (Manually, usually done by Ursual or I). Next time an OOPS
summary is generated, it'll know about the bug report and add a link
with the aggregate OOPSes of the same type.

OOPSes of the same type are identified by a signature, which is
computed from the exception type and exception value.

Does the above make things clearer about oops-tools? Would it be able
to address your use case?


[1] http://launchpad.net/oops-tools

>>
>> > Perhaps
>> > a weekly email here with counts of classified tracebacks would
>> > help us reduce the number?
>> >
>>
>> Hmm. I think I see. We already do something like this for other
>> subsystems of Launchpad, but I think that's mostly based on the OOPS
>> codes, rather than tracebacks.
>>
>> I'm CCing Diogo & Ursula, who know way more about this than I do.
>
> I took inspiration from the code in apport. Only slight modifications
> were needed for something that is simple and seems to work quite well.
> I put it all together in to a cron script, and I'll send an annotated
> version out in a minute.

Can you send me the annotated version too? That way I'll be able to
understand better your use case.

Cheers,
-- 
Diogo M. Matsubara