[Lxc-users] lxc-instance fail

[Serge Hallyn]

Quoting Tamas Papp (tompos at martos.bme.hu):
> 
> On 10/04/2013 04:11 PM, Tamas Papp wrote:
> > hi,
> >
> > Starting a container from cli by lxc-start works fine, lxc-info shows the right state.
> >
> > Using upstart:
> >
> > start lxc-instance NAME=archiva-test CONFIG=/tank/lxc/archiva-test/config
> >
> > # lxc-info -n archiva-test
> > state:     STOPPED
> >
> > Though the container gets started.
> >
> > As a result I cannot use lxc-stop, lxc-ls or any other tool.
> > But, stop lxc-instance... works fine.
> >
> >
> > I was able to reproduce this issue on various machines with the latest daily build.
> >
> >
> > Am I alone with this issue?
> 
> This issue is bigger problem then I thought.
> If container is started with the upstart script, a second one can be
> started via normal lxc-start.
> 
> Why is that?
> I mean, lxc tools doesn't see the container itself for some reasong.
> But should not the directory be locked by the previously running
> container? Eg. though the rootfs.hold file.
> 
> It's quite a big issue, that two instance is able to run on the same
> directory.

An easy way to reproduce this without zfs is:

	sudo lxc-create -t ubuntu -n r0
	sudo cp /var/lib/lxc/r0/config /tmp
	sudo lxc-start -n r0 -f /tmp/config -d
	sudo lxc-info -n r0
		(it shows up as stopped)

sudo lsof shows that

	@/var/lib/lxc_anon/r0/command

is the command socket, not the usual

	@/var/lib/lxc/r0/command

You can see it using

	sudo lxc-info -P /var/lib/lxc_anon -n r0

Perhaps lxc-instance should be following the config link, making
sure it is in fact $lxcpath/$lxcname/config, and then just doing
lxc-start -n $lxcname?

-serge