VirtualDomain status waits forever if libvirtd is not running

[dmuhamedagic]

If the libvirtd is down or virsh cannot connect to it for whatever reason, VirtualDomain considers that as "no state" for the domain. That turns it into an infinite loop and operations eventually time out (stop too). Noticed while shutting down a node where libvirt got erroneously stopped before pacemaker: the node got fenced because a VirtualDomain resource timed out first in migrate then in stop.

[dmuhamedagic]

It would be good to add test whether libvirtd is operational in the validate procedure and then exit with ERR_INSTALLED if not (or NOT_RUNNING on stop). It seems like it'd be enough just to query the version: virsh -c ... version

[drstevenhale]

I run libvirtd as a resource and then add it as an ordering constraint on my VMs.

[dmuhamedagic]

On Fri, Jan 16, 2015 at 02:09:07AM -0800, stevenhale wrote:

I run libvirtd as a resource and then add it as an ordering constraint on my VMs.

Right, that would obviously help. I'm really not sure why, but
most of the configurations I've seen don't do that, but let the
init system start libvirtd on boot. Perhaps to run guests which
are not controlled by the cluster?

[davidvossel]

Virtualdomain can detect the status of qemu and lxc domains without libvirt being present. We look for the process id.