vzctl --wait problem [message #12815] |
Thu, 10 May 2007 20:08 |
sodi
Messages: 6 Registered: May 2007
|
Junior Member |
|
|
Hi,
Sometimes
vzctl start 101 --wait 3
results in a kernel panic of the host, such that I have to reset the whole server.
This is not deterministic, sometimes it happens already on the first start and sometimes after some starts and stops.
It never happens without the wait option. As far as I can see there is some usleep() call which causes the crash. So this might also be a problem with the linux kernel of the host
Any ideas ? Why ? How to solve ?
I compiled a linux 2.6.18 kernel with current patches. If the start works everything is fine.
The machine is a 8 core x86.
And yes, the --wait parameter is important since we want to execute some scripts after starting the ve, so I need a running linux.
Thanks for any help or hints.
|
|
|
|
|
|
Re: vzctl --wait problem [message #12864 is a reply to message #12835] |
Sat, 12 May 2007 18:37 |
sodi
Messages: 6 Registered: May 2007
|
Junior Member |
|
|
Thanks for the serial console/netconsole hints, I think this would be some last chance try (netconsole might work for me)
Also since the installation usual is done remotely and the KVM is not always on, it's only by chance to get a photo from the console.
Anyway, after changing the HPET support in kernel compile options to "no" and recompiling the kernel, I didn't get the kernel panic any more. May be this is a specific problem to a 2 Quad-Core Intel Xeon system. I don't think this is really an openVZ problem, but a general kernel problem.
(It was the same .config except the HPET was disabled)
Since then I created several ve, stopped / started them w/o any problems.
Sorry, that I can't be more helpful with console information. I know that if there is a problem, it could generally improve openVZ, if the reason is known.
[Updated on: Sat, 12 May 2007 18:38] Report message to a moderator
|
|
|
|
Re: vzctl --wait problem [message #12905 is a reply to message #12815] |
Mon, 14 May 2007 16:21 |
sodi
Messages: 6 Registered: May 2007
|
Junior Member |
|
|
The configuration for the "working" kernel is appended (I didn't change most of the default options). I assume at least that it's working, since I didn't have any problems with it by now. With HPET enabled the crash happened about once in ten trials.
As template I used "opensuse-10-i386-default.tar.gz", since the machine is not allowed to talk to web, I used the precreated template. But still I don't think this is really a problem of the template, I didn't change it.
I was guessing for HPET, since it seemed to me that there had been several patches applied between .18 to .21 wrt SMP. I'm not a kernel expert, just looking for patches for timer / HPET.
|
|
|