How ABRT works

When your application crashes, the crash event is handled by one of ABRT’s language or runtime dependent hooks which forwards the crash information to abrt daemon. The daemon stores the data in /var/spool/abrt and runs a chain of events that for example collect more data about the system, produce backtrace from core dump or notifies a user. This process is followed by automatic or manual reporting to various supported targets like bugzilla or ABRT Analytics (ABRT Analytics).

Reporting

Currently, typical desktop reporting work-flow consists of generating so called μReport (micro report) designed to be completely anonymous so it can be sent and processed automatically avoiding costly bugzilla queries and manpower.

The ABRT Analytics (also known as ABRT server in the past) in this scenario works like a first line of defense — collecting massive amounts of similar reports and responding with tracker URLs in case of known problems.

If user is lucky enough to hit a unique issue not known by ABRT Analytics, reporting chain continues with reporting to bugzilla, more complex process which requires user having a bugzilla account and going through numerous steps including verification that the report doesn’t contain sensitive data.

ABRT also features unattended reporting possibilities in form of email reports, upload functionality and integration with tools like Spacewalk or Foreman for large scale deployment scenarios.

Components

abrt

The daemon and collection of tools for handling crashes and monitoring logs for errors. Takes care of storing and processing of the crashes caught by various hooks.

gnome-abrt

Simple GUI application for problem management and reporting.

_images/gnome_abrt.png

libreport

Libraries providing an API for reporting problems via different paths like email, bugzilla, ABRT Analytics, scp upload..

ABRT Analytics

Crash collecting server, also known as ABRT Analytics (or ABRT server in the past). Provides accurate statistics of incoming reports and acts as a proxy in front of bugzilla (or any other issue tracker) when it comes to automatic reporting of crashes. It’s designed to receive anonymous μReports and to find clusters of similar reports among them. For reports that are known, user receives fast response containing links to ABRT Analytics’s problem page, issue tracker or an entry from knowledge base, that contains number of well-known issues like usage of proprietary kernel modules or browser crashes caused by unsupported modules.

satyr

Satyr is a collection of low-level algorithms for program failure processing, analysis, and reporting supporting kernel space, user space, Python, and Java programs. Considering failure processing, it allows to parse failure description from various sources such as GDB-created stack traces, Python stack traces with a description of uncaught exception, and kernel oops message. Infromation can also be extracted from the core dumps of unexpectedly terminated user space processes and from the machine executable code of binaries. Considering failure analysis, the stack traces of failed processes can be normalized, trimmed, and compared. Clusters of similar stack traces can be calculated. In multi-threaded stack traces, the threads that caused the failure can be discovered. Considering failure reporting, the library can generate a failure report in a well-specified format, and the report can be sent to a remote machine.

retrace server

The retrace server provides a core dump analysis and backtrace generation service over a network using HTTP protocol.

Privacy

μReports consist of structured data suitable to be analyzed in a fully automated manner, though they do not necessarily contain sufficient information to fix the underlying problem. The reports are designed not to contain any potentially sensitive data to eliminate the need for review before submission.