Eric Sandeen writes:
Sam Varshavchik wrote:
> nodata writes:
>
>> Am Freitag, den 12.12.2008, 17:57 -0500 schrieb Sam Varshavchik:
>>> I've got a quad-core dual x86_64 server that, so far, reproducably locks
up
>>> under every 2.6.27 kernel released for F9 so far, when I run a build cycle.
>>> The last kernel that manages to survive under load is 2.6.26.6-79.fc9, so
>>> I'll continue to boot it until I get a 2.6.27 kernel that doesn't
croak on
>>> me.
>>>
>>> Given that the hardware does not have a serial port, how else can I capture
>>> an oops, if one is being generated? At least I hope that there'll be an
oops
>>> for me to capture.
>>>
>> Take a look at netdump.
>
> Took a look.
>
> 1) Only the server component is present in Fedora. There is no client. And,
> of course, I need the client.
I haven't used netdump for a while but:
http://kojipkgs.fedoraproject.org/packages/netdump/0.7.16/13/x86_64/
for example seems to have both. *shrug*
> 2) The client package requires additional kernel modules to be built. If you
> are actually using netdump in Fedora, please explain how.
>
> 3) netdump only supports a small set of NICs. netdump does not support my
> NIC.
anyway, how about kexec/kdump/crash? Maybe with a little sysrq-C thrown
in if it's not actually panicing?
Well, I boned up on the kexec/kdump/crash tools.
Immediately found a whole bunch of bugs in mkdumprd that produces a totally
useless crash initrd image for me. I was able to fix some of the problems,
but still don't have a workable crash image. Working on it.
Is anybody using kexec-tools on a *freshly* installed F9 (not an upgrade
from earlier Fedoras) that uses RAID volumes? Did it work for you out of the
box? As far as I can tell, it can't, because Anaconda no longer creates
modules.conf for new installs, but mkinitrd figures out that the regular
kernel boot images needs scsi_mod, sd_mod, and your hardware IDE/SATA/SCSI
module, yet mkdumprd still looks for "alias scsi_hostadapter" in
modules.conf, and without it you get a dumprd image without the necessary
modules need to load scsi support. Even after I fixed that, I still don't
get RAID to come up in the crash initrd image, not quite sure why.
Bug 476368.