[Lustre-discuss] debugging kernel panics

Klaus Steden klaus.steden at thomson.net
Mon Nov 12 13:01:47 PST 2007


Hi folks,

I'm running Lustre with the 2.6.9-55.0.2.EL_lustre.1.6.2smp kernel, and one
of the systems in my cluster keeps crashing repeatedly, at irregular
intervals.

I've been saving the core files (or at least partial core files) by using
'netdump', but in order to diagnose the cause, I think I need a kernel with
debugging symbols in it to make sense of the core image.

Does one exist on the Lustre download site? Is it trivial enough to build
one? The last time I built a kernel from SRPM, it tagged it with a date/time
stamp, which might introduce variation in the kernel image and produce a
mismatch between kernel and core image.

The only files I see in my /boot directory are vmlinuz compressed kernel
images -- is there any way to unpack them to examine them for debugging
symbols?

What's the usual procedure in a case like this?

thanks,
Klaus




More information about the lustre-discuss mailing list