[CRIU] [Lxc-users] Do we have a plan to make the checkpoint/restart feature.

Serge Hallyn serge.hallyn at ubuntu.com
Fri Dec 6 13:00:05 PST 2013


Quoting Pavel Emelyanov (xemul at parallels.com):
> On 12/04/2013 10:12 PM, Serge Hallyn wrote:
> > Quoting Nipun Arora (nipun at cs.columbia.edu):
> >> Hi,
> >>
> >> I've been tryint criu for checkpointing and restart. One of the issues is
> >> that criu does not support external console.
> >> I tried setting lxc.console = none, and lxc.tty = 0... but seem to get no
> >> progress when I run the following command..
> >>
> >> lxc-create -t ubuntu -n test_lxc -f test_config
> >> lxc-start -n test_lxc
> >>
> >> The config file basically has only the following 2 lines
> >> lxc.console = none
> >> lxc.tty = 0
> >>
> >> Can someone help out in explaining why this is happening?
> > 
> > I think it's been explained before - the external console points
> > to a device file sitting in another mount namespace.  criu can't
> > checkpoint that.
> 
> Ack.
> 
> > Back in the in-kernel checkpoint days, IIRC we allowed such files
> > to be marked as uncheckpointable in the image, and the restart
> > program to fill those fds in with fds it had opened beforehand.
> > 
> > Can that be done with crui?
> 
> In theory -- yes. IOW -- we need to patch criu to achieve that. I
> planned to make this support using plugins to criu that help dumping
> external mount points. Currently Andrey works on such a plugin for 
> external unix sockets, you can find his patches on the list From: avagin at . 

sorry, which list?  :)

> Similar thing can be done for external (bind)mounts. Does it make
> sense for you?

Yes, that sounds good.  And a fast criu-image-grep util which can tell
the caller (lxc-restore) which fds and mounts and maybe nics the
image needs filled in.

-serge


More information about the CRIU mailing list