svn commit: r197065 - in stable/8: etc/defaults lib/libc/stdlib sys/amd64/conf sys/i386/conf sys/ia64/conf sys/pc98/conf sys/powerpc/conf sys/sparc64/conf

[John Baldwin]

On Friday 11 September 2009 2:00:58 am Jason J. Hellenthal wrote:
> On Thu, 10 Sep 2009 13:00 -0000, rwatson wrote:
> 
> >
> > On Thu, 10 Sep 2009, Remko Lodder wrote:
> >
> >> I agree with that; it would (!) help the bugbusting team in gathering 
> >> required information. If there is an way to automate crashdumps and proper 
> >> reporting and stick that in /var/crash/crash.$date or something and tell 
> >> people that they can report their problems on the bugs list where needed, 
> >> we have information upfront. I still remember the time where we had to 
> >> chase people to get this information, sometimes never being able to 
> >> properly get the information. If it is there by default, it will help.
> >> 
> >> Please consider keeping it enabled..
> >
> > Well, John's crashinfo(8) tool from Yahoo! does all this, the only "problem" 
> > is that it requires extracting the crashdump to a local file system.  It 
> > seems like we have a lot of the right ingredients, maybe we can have a 
> > brain-storming session on distributed diagnostics and debugging at the 
> > Devsummit on how to find some better middle grounds here.
> >
> > Robert N M Watson
> > Computer Laboratory
> > University of Cambridge
> > 
> >
> 
> If I may, I would like to introduce a distributed targeting system to this 
> conversation in addition to crashinfo. Given with the above conversations I cant 
> help but think that in a case like this it would be helpful to setup a central 
> database for collection of information and write a little bit more code into 
> crashinfo for uuencoding a blob to send through email or maybe another way so 
> data can be collected, sorted & analyzed with statistics spilled out into a web 
> page for review.

My previous job actually stored the output of a crashinfo-like script for
each panic into a database along with metadata such as a one-line summary
(it even had some limited ability to auto-tag known panics).  However, it
required a good chunk of time *every* day to check new panics received each
day to see if any were not previously seen before, or new variants on an old
panic, etc.  The time was non-trivial and I do not think would scale well to
the all-FreeBSD-users universe.

-- 
John Baldwin