Open Stack

On 16 April 2015 at 18:11, Anthony PERARD <anthony.perard at citrix.com> wrote:
> Hi all,
>
> With libvirt_type=xen, the tempest test
> tempest.scenario.test_volume_boot_pattern.TestVolumeBootPattern.test_volume_boot_pattern
> fail.
>
> This is because it setup a volume with 'vda' as device_name and nova (well
> libvirt) fail to bring up an instance. 'vda' (virtio) is not supported by
> libvirt-xen.
>
> I tried to rename 'vda' to 'xvda' in the test, but then nova fail earlier.
> When nova is setting up a block device list for the instance, it set a
> value for boot_index. boot_index is set to 0 only if the device_name is
> equal to root_device_name, which default to 'vda'.
> The only way too change root_device_name is to add it to the metadata of
> the image but I don't know if that a good idee.
>
> How could we resolve this? Skip the test if libvirt_type=xen?
>
> I have started to write a bug report for this:
> https://bugs.launchpad.net/tempest/+bug/1443898

Hi,

This on a bit of the API which feels too leaky around the
virtualisation driver chosen :(

We do have a horrific hack in place so it works with the XenAPI here:
https://github.com/openstack/nova/blob/master/nova/compute/utils.py#L136

I guess its possible using that hack and tweaking the libvirt code
could allow you to move forward with that.

Longer term, we need a better way of dealing with these differences
within the BDM code. Its possible there is something for this that I
don't know about.

Thanks,
John