On 01/14/2013 04:19 PM, Vincent Veyron wrote:
> The only downtime I had in over two years was due a forced bios upgrade
> by the hosting service, and I have no formal training in server
> administration. Debian stable certainly works.
My personal server is on Debian too, with a similar uptime. But we
recently ran into this guy on our 12.04 Ubuntu systems:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1055222
I sent a message to [Performance] a while back suggesting disabling this
setting, and it's still true. But apparently if you set
sched_autogroup_enabled to 0 via sysctl, you won't be able to boot
because the kernel will panic before it finishes. Using the setting at
all makes reboots crash. And apparently, some tasks can occasionally get
into a bad state such that deallocating them causes a 0000 kernel oops
in set_task_cpu. That's hilariously bad, and all due to a merge from
upstream from a few months ago.
I probably wrongly attributed this to Ubuntu since it's the top result
in searching for 'sched_autogroup_enabled panic', but the real point was
that there doesn't seem to be any official support. We couldn't escalate
this problem to anyone anywhere, except by tagging the bug report or
opening our own.
Even calling canonical to ask about buying a support contract got us an
automated "We'll contact you within two business days" response, which
isn't exactly ideal. So we're strongly considering RHEL, because at
least they would call us back, and would give us some small amount of
peace knowing we could maybe get some assistance since we don't exactly
have a kernel dev on staff to find things like this.
We did figure it out eventually, but it took a couple hours to figure
out why the machines wouldn't boot, and all weekend to fully isolate the
problem as the cause of the other crashes.
So of course, we had to ask. I know this could have (and has, in my
experience) happen to any distro, but it just seems more... prevalent in
Ubuntu.
Then again, our older RHEL systems crashed like it was their job when we
were using our onboard Broadcom NICs. Maybe we're expecting too much.
--
Shaun Thomas
OptionsHouse | 141 W. Jackson Blvd. | Suite 500 | Chicago IL, 60604
312-676-8870
sthomas@optionshouse.com
______________________________________________
See http://www.peak6.com/email_disclaimer/ for terms and conditions related to this email
