ABRT Server

[Michal Toman]

Greetings everybody!

As you may have noticed, new ABRT 2.0.12 has recently been pushed to 
Fedora. Besides significant changes in reporting workflow, it also uses 
the ABRT server to collect what we call uReports (micro-reports).

uReport is a short (usually < 4kB) fingerprint of the crash, with no 
private information in it. It contains some basic data (package, 
architecture, kernel... NO hostname, NO username etc.), together with a 
debuginfo-less backtrace consisting of binary names, build-ids, function 
offsets, function names (if available) and some fingerprints helping in 
duplicate detection.

After uploading the uReport to ABRT server, the backtrace is completed 
- missing function names, source file names and line numbers are added. 
A clustering algorithm is ran periodically on the reports. It searches 
for similar uReports and groups them into "Problems". We divide these 
into two categories:
1) Hot problems: started appearing recently and happen often. Should be 
fixed ASAP.
2) Long-term problems: started appearing long time ago, they do not 
happen often, but they are still reported sometimes. There are no 
long-term problems yet, as there is some minimal age limit.

After receiving uReport, the server responds whether the problem is 
already know or not. In the case it is known, the reporting process 
stops and you can see both Bugzilla URL and ABRT Server URL. Otherwise 
standard BZ-reporting process continues.

You can browse reports and problems through the webUI
https://retrace.fedoraproject.org/faf/summary/

We believe this will help developers to better prioritize their work and 
make debugging easier (crashes in common libraries are grouped into a 
single problem, for each crash versions of affected 
packages/architectures are listed etc.).

Any feedback or RFEs are welcome [1]!
Michal & ABRT Team

[1] https://fedorahosted.org/abrt/