Although OpenStack XenAPI supports some other disk formats, e.g. ami, raw, iso; VHD is the most commonly used and strongly recommended disk format. By comparing to raw disk, the occupied physical size can be much smaller than the virtual size with VHD. Actually VHD is the default disk format in XenServer OpenStack. This blog will focus on how to create images basing on VHD disk format; how to convert other types of disk format to VHD and finally how to create Windows images for XenServer in OpenStack.

Generate Images from VHD images

For XenServer’s OpenStack integration, the VHD disks should be contained in gzipped tarball, as we can support having multiple differencing disks (snapshots) from a single base VHD. And the VHD file itself should be created by vhd-util(having the originator set to “tap”), otherwise it will fail at resizing at booting VMs from the image. So it’s recommended to create a VDI on a EXT SR in the XenServer and import the source VHD file into this new VDI. And then we can create OpenStack image basing on new VHD file.

For example, there is a zipped VHD-based Ubuntu image at:

https://cloud-images.ubuntu.com/vivid/current/vivid-server-cloudimg-amd64-disk1.vhd.zip

Let’s see how to generate images from it.

  • Firstly download image and extract vhd file from it.
  • wget https://cloud-images.ubuntu.com/vivid/current/vivid-server-cloudimg-amd64-disk1.vhd.zip
    unzip vivid-server-cloudimg-amd64-disk1.vhd.zip
  • On a XenServer which has a EXT SR(e.g. uuid=55654811-fa04-ba61-6c11-59116f85399f), create a new VDI and import this VHD file to it:
  • [root@baras tmp]# vSize=$(($(vhd-util query  -n vivid-server-cloudimg-amd64-disk1.vhd -v) * 1024 * 1024))
    [root@baras tmp]# xe vdi-create sr-uuid=55654811-fa04-ba61-6c11-59116f85399f name-label=tmpVDI type=user virtual-size=$vSize
    a4bb1d38-5026-47ed-a400-b5fd205c5339
    [root@baras tmp]# xe vdi-import uuid=a4bb1d38-5026-47ed-a400-b5fd205c5339 filename=vivid-server-cloudimg-amd64-disk1.vhd format=vhd
  • Under the SR mount path, you will see a VHD file named as <VDI-uuid>.vhd. Let’s copy this file to another path:
    [root@baras tmp]# ls /var/run/sr-mount/55654811-fa04-ba61-6c11-59116f85399f/a4bb1d38-5026-47ed-a400-b5fd205c5339.vhd
    /var/run/sr-mount/55654811-fa04-ba61-6c11-59116f85399f/a4bb1d38-5026-47ed-a400-b5fd205c5339.vhd
    [root@baras tmp]# cp /var/run/sr-mount/55654811-fa04-ba61-6c11-59116f85399f/a4bb1d38-5026-47ed-a400-b5fd205c5339.vhd ./
  • Rename vhd file and create gzipped tarball
    [root@baras tmp]# mv a4bb1d38-5026-47ed-a400-b5fd205c5339.vhd 0.vhd
    [root@baras tmp]# tar -czf vivid-server-cloudimg-amd64-disk1.tgz 0.vhd
    
  • Create image and import data to glanceglance image-create --name="Ubuntu-vivid-server-cloudimg-amd64" --is-public=true --container-format=ovf --disk-format=vhd --property vm_mode=hvm --file vivid-server-cloudimg-amd64-disk1.tgz

Converting an existing QCOW2 image

Currently, XenServer doesn’t support QCOW2 images, but thankfully it’s easy to convert QCOW2 to VHD-based images by using qemu-img. We can download QCOW2 images from:

https://getfedora.org/cloud/download/

https://uec-images.ubuntu.com/releases

For example, we download Fedora qcow2 formatted image. We can get the vhd file via the the following command:qemu-img convert -O vpc Fedora-Cloud-Base-23-20151030.x86_64.qcow2 Fedora-Cloud-Base-23-20151030.x86_64.vhd

Then we can follow up the similar steps described in the section of “Generate Images from VHD images” to create OpenStack images from the above VHD file.

Here, I only take QCOW2 as the example, logically we can create XenServer images from any other images format as long as the images can be converted to VHD.

Creating from an existing XenServer VM

If we have an existing VM running on XenServer, it’s easy to create the image from this VM. But please ensure you have cloud-init installed and configured correctly as OpenStack depends on cloud-init to initialize VMs at first boot.

1. shutdown the VM

2. get VM’s VDI and export it as VHD file. Usually that’s the first disk of the VM, so at here I use “device=xvda” to ensure only the first VBD is listed. If that’s not the first one, please identify it by yourself and specify the correct device.

vbd_uuid=$(xe vbd-list vm-name-label=${vm-name} device=xvda minimal=true)
vdi_uuid=$(xe vdi-list vbd-uuids=${vbd_uuid} minimal=true)

3. Using the vdi_uuid to find the VHD file under the SR mount path, in case this is an EXT SR; otherwise you should export this VDI to file system and then follow the steps in the section of “Generate Images from VHD images” to import this VHD to an EXT SR’s VDI and take that VDI’s VHD file directly.

4. Once we get the VHD file, we can follow the same way described in the section of “Generate Images from VHD images” to generate the OpenStack image.

Creating a Windows image

In this section, I will try to describe how to create a Windows Image on XenServer. Please note this should be done on XenServer 6.5 or later and have all hotfixes installed (particularly the PV tools hotfixes). The most recent PV tools will bind to multiple device_ids (see step 5 below), and therefore do not require special flags to be added to the image to set the device ID when creating a new VM.

1. From XenCenter, create a VM. When OpenStack creates a VM it does not have the per-distribution settings in the Template, so “Other install media” has a closer set of options to that presented by OpenStack. So at here we suggest you choose the option of “Other install media” which will give a default HVM template.win10-install-1

Select the CD image to install from. In this example, we are installing Windows 10 (64-bit)

win10-install-2

2. Use Windows Updates to install the latest updates. If you’re using the Enterprise version of XenServer 7.0, then the PV drivers should also be automatically downloaded and installed for you (auto updating when new versions are released as well), so you can skip to step 6.

3. Install PV driver from XenCenter:

PVdriver-1

4. Usually the above operation will insert the XS tools DVD into the DVD drive and finish the installation automatically; but for windows you may need to manually install the XS tools from the VM.

PVdriver-2

5. Check that the device_id has not been set by the XenServer tools.

Some versions of the PV drivers may set a device_id to the VM’s parameter of platform, even if it is not specified on the template. If a device_id is set on the VM during installation, OpenStack must have the proper device_id in the image’s metadata so that the PV driver could bind xenbus driver to the correct platform device after the guest VM booted from this image. \ In order to avoid potential issues, let’s ensure this parameter is not set (Actually with choosing “Other install media” as the template in step 1, it should be unset).xe vm-param-list uuid=8f151bb3-4c90-3e65-94b3-5a9602380369 | grep device_idIf device_id is set, you can use the following command to unset it, but you may also need to re-install the PV drivers.
xe vm-param-remove param-name=platform param-key=device_id uuid=8f151bb3-4c90-3e65-94b3-5a9602380369

6. Add another Administrators account: The Window 10’s security strategy doesn’t allow the built-in Administrator to run Microsoft Edge which will be used to download cloudbase-init. The built-in Administrator will be hidden per Windows’s security strategy.

7. After login with the new account (e.g. myadmin). Downloading and install the latest Cloudbase-Init version: https://cloudbase.it/downloads/CloudbaseInitSetup_Stable_x64.msi

cloudbase-init-1

cloudbase-init-2

8. After cloudbase-init finished running Sysprep, it will shut down VM.

9. Get the VHD file from XenServer:

[root@baras tmp]# vbd=$(xe vbd-list vm-name-label=Win10-VM device=xvda --minimal)
[root@baras tmp]# vdi=$(xe vdi-list vbd-uuids=$vbd --minimal)
[root@baras tmp]# ls /var/run/sr-mount/55654811-fa04-ba61-6c11-59116f85399f/$vdi.vhd
/var/run/sr-mount/55654811-fa04-ba61-6c11-59116f85399f/cbefbad5-804f-40f3-ac5c-fe8f2e279110.vhd
[root@baras tmp]# cp /var/run/sr-mount/55654811-fa04-ba61-6c11-59116f85399f/cbefbad5-804f-40f3-ac5c-fe8f2e279110.vhd ./0.vhd

10. Create image and upload to glance:

tar -cvzf win10.tgz 0.vhd
glance image-create --name="win10" --visibility=public --container-format=ovf --disk-format=vhd --property vm_mode=hvm --file win10.tgz)

11. Create a new VM from the new image to verify this image.nova boot --flavor 3 --image win10 --meta admin_pass=testVM1pass --nic net-id=0cb33381-e48b-444a-8709-fde15d4cab4e Win10-testVM1

12. Login Windows with myadmin by using password as the one specified by “–meta admin_pass=”; check this VM’s hostname is Win10-testVM1 which is the VM name specified in the nova boot command.

PV Guests

XenServer does, of course, support PV guests under OpenStack, and generation of these images is much the same as for HVM guests. The key difference is when uploading them to glance, specify “–property vm_mode=xen” instead of hvm.

Give it a go

XenServer-based OpenStack clouds are really easy to deploy using Mirantis OpenStack and the XenServer Fuel plugin – check out our other blog posts like Introduction to the XenServer Fuel Plugin orDeploying Mirantis OpenStack on a single XenServer – so it’s never been easier to create and test your own XenServer OpenStack images.

Also, do check out XenServer 7.0 enterprise edition to get features like the Automated Windows VM Driver Updates to make your cloud easy to maintain.

BLOG BANNER