First Failure Data Capture Using Crash Dumps

First Failure Data Capture (FFDC) is now one of the most important requirements for Linux OS in deployment. Given the rapid pace at which Linux is getting deployed on servers and enterprise systems, customer support is now of prime importance. FFDC, no doubt, plays a critical role in both problem identification and early problem resolution.

Many a time, customer problems are hard to recreate. Further, it occurs very infrequently at customer sites thus making it almost impossible to devise a debug strategy. In such cases, effective FFDC is the only way out. Linux currently supports various tools that offer FFDC capability to the kernel. The Linux Kernel Crash Dumps (LKCD) project, the RedHat netdump project, the traditional OOPS messaging scheme, kmsgdump, all provide mechanisms of collecting failure related data for analysis.

Crash dumps continue to remain one of the most effective solutions for first failure data capture. Though no one solution has made it to the mainline kernel, there exists several projects that provide crash dump facilities. The aim of this BOF is to discuss the state of crash dumping tools in Linux, the directions that the various projects are taking and their future.

Some initial topics of discussion for this BoF are

- Can Linux still continue to do without a mainline crash dumping tool?

- What are the benefits of using crash dumping tools, over traditional Linux tools such as oops?

- Where do the existing crash dumping tools stand today?

- How wide (or narrow) are their usage?

- What additional capabilities can be provided?

- What can get a crash dump project into the mainline kernel?

- What are the mistakes that we made during the 2.5/6 development cycle?

- What do kernel developers think, in terms of the most effective crash dumping technique that can be used? Is kexec-based-dumping really the way to go?

- Where do we stand in terms of compatibility between various dump analysis tools (crash,lcrash,gdb,etc)?