Open Stack

Hello Colleagues,

I have made some progress with passing through a 'raw' scsi disk from an openstack hypervisor
to one of its guests. However, I've now hit a libvirt problem and am wondering whether others may
have dealt with this already in the ubuntu world (I believe it is resolved in the centos world).

On a (recently built) openstack-ansible 15.1.6 environment where one compute node is
also a storage node I do the following:

1. define an image of xenial-server with the following properties set:

      hw_disk_bus='scsi'
      hw_scsi_model='virtio-scsi'

2. create a volume on a given compute node:

      # openstack volume create --size 269 --availability-zone special v0
      # openstack volume show v0 -f json | jq .id
      "6dc7de73-7d5b-4410-8c62-a06e722187e9"

3. create an instance on the same compute node:

      # nova boot --flavor xenial-server --image xenial --key-name test-key \
          --nic net-id=0395285e-b548-458b-9db5-f8d869e49a06 --availability-zone nova:compute1 \
          --block-device id=6dc7de73-7d5b-4410-8c62-a06e722187e9,source=volume,dest=volume,bus=scsi,type=lun \
          xenial0

4. Step 3 attempts to build the instance but errors out with the following in the
   nova-compute.log on compute1:

      [instance: 97b87f5e-b146-4c40-8128-44446c996b8e] libvirtError: unsupported configuration: \
        scsi-block 'lun' devices do not support the serial property

    (nb: this is the expected instance id)

It seems that this is a bug in libvirt that is fixed in the centos/rhel world.

The hypervisor 'compute1' is built with 16.04.2 LTS and fully updated.

Does anyone know of a fix in the ubuntu/debian world?

Or should I be raising this on a xenial forum?

Kind regards, Lawrence

Lawrence J Albinson
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20170718/33644712/attachment.html>