[Lxc-users] cannot start any more any container?! (partially solved)

Daniel Lezcano daniel.lezcano at free.fr
Fri Oct 21 13:00:05 UTC 2011


On 10/21/2011 02:33 PM, Ulli Horlacher wrote:
> On Thu 2011-10-20 (09:00), Papp Tamas wrote:
>
>> Well, I don't see anything like this. Actually I use 0.7.5. Try to upgrade.
> After upgrading to lxc 0.7.5 the problem is still there:
> I cannot start any container and there is no (log) output at all. For
> every lxc-start command I get a new veth interface and the lxc-start
> process is not killable (uninterruptable waiting for IO).
>
> At this point I gave up and tried the Windows problem solving methode:
> rebooting (the host server).
>
> After reboot, I can start and stop containers without any problems.
> Everything works fine, as it should.
>
> I am not happy with this state: I do not know what went wrong and I have
> no solution if this problems reappears, besides rebooting, which will
> terminate all other container VMs, too. This is a NO-GO for a production
> environment!
>
> I have now installed linux-image-server-lts-backport-natty (Linux 2.6.38)
> and hope (*) this fixes the bug.
>
>
> (*) Hope and faith belongs to the church and not to a computing centre.

The problem you are describing is not related to LXC but to the network
namespace where a dangling reference in the kernel with ipv6 locks the
network devices. When the kernel hits this bug, any process creating a
network device or deleting one will be stuck in an uninterruptible state.

If you are able to start a container with an ipv6 address
(lxc.network.ipv6=xxx), stop it, and start it again 10 seconds later
then that means the bug is solved in the kernel.

The key point is what Serge said, if you have this message in your console:

"kernel: unregister_netdevice: waiting for ... to become free"

then this is a kernel bug.

If you still have this problem with 2.6.38, please let me know, I will
ping the netdev@ to fix it and propose it as a fix for natty server
ASAP, so that won't be a no-go problem for you.

Thanks
  -- Daniel






More information about the lxc-users mailing list