[Lxc-users] Ubuntu template questions

[Fajar A. Nugraha]

On Thu, Mar 15, 2012 at 10:02 AM, Serge Hallyn
<serge.hallyn at canonical.com> wrote:
> On 03/14/2012 03:23 AM, Fajar A. Nugraha wrote:
>> On Wed, Mar 14, 2012 at 3:17 PM, Jäkel, Guido<G.Jaekel at dnb.de>  wrote:
>>> Dear Fajar,
>>>
>>> i just googled  http://www.makelinux.net/man/7/P/power-status-changed  . There's written:
>>>
>>>         This event is not handled in the default Upstart configuration.
>>>
>>>
>>> For control-alt-delete, the corresponding sentence states:
>>>
>>>         In the default Upstart configuration handling of this event is provided by the /etc/init/control-alt-delete.conf task which runs the shutdown(8) tool.
>>>
>>>
>>> This sounds to me like in the current version of upstart the suggested patches to add a SIGPWR handler are included and there's "just" a script missing. I don't have a Ubuntu available; maybe a simple  power-status-changed.conf  will already do all the magic???
>>
>> Good catch :D
>>
>> $ cat power-status-changed.conf
>> start on power-status-changed
>>
>> task
>> exec shutdown -h now "Power Down"
>>
>>
>> That, plus an lxc-ps and kill -PWR from the host, was able to shutdown
>> the guest cleanly with minimal change to the guest.
>>
>
> Thanks guys, this is great.  I don't know if we can swing this this
> cxycle (it's possible) but an upstart package with your job added is at
> https://code.launchpad.net/~serge-hallyn/ubuntu/precise/upstart/upstart-handle-sigpwr


Did you also have a chance to modify /etc/init/lxc.conf?

A good shutdown script would probably goes something like this:
- list all containers to autoshutdown.
The existing one looks at /etc/lxc/auto/*, but IMHO it might be better
to just list ALL runing containers using "lxc-ls", since they're going
to be dead anyway when the host is stopped.
- get PID of init process in containers to shutdown. A combination of
"lxc-ls", "grep", and/or "awk" would probably work. If the pid can't
be determined, shutdown the container immediately using lxc-stop
- send SIGPWR to all init PIDs in previous step
- create a wait loop for a maximum of ... 30 seconds (?) which
basically check whether all the process with PIDs above still exist or
not. If it STILL exist at the end of the wait time, we assume the
container can't be shutdown cleanly.
- force-shutdown containers whose init PID hasn't disappear yet using lxc-stop.

The maximum wait time is debatable, but IMHO 30 seconds should be a
good start. Or perhaps we need to put it in /etc/default/lxc?

-- 
Fajar