My company is migrating to a new infrastructure for its virtual machines and switching from XenServer to KVM. This week we had to tackle the migration of existing Xen-VMs to the new hypervisor – turned out that this has been trickier than I thought it would be!
This is a description of how we migrate VMs from a XenServer to an Ubuntu 14.10 with KVM. The VMs run with Debian.
Get the image to the destination server
You have to get an image from the Xen VM. Our machines are snapshotted every night (using NAUBackup), so for my test vm "TestVM01" I already had the appropriate .xva image.
The next step is to transfer that image to the destination server running KVM. Before you transfer the image you might want to zip it first:
tar czvf TestVM01.xva.tar.gz /path/to/TestVM01.xva
Then copy it to the KVM-server. I've used
scp for that task:
scp xenserver.example.com:/path/to/TestVM01.xva.tar.gz ~/xen-migration/ # from the destination server
Switch to the download directory and untarzip the whole thing:
tar xzvf TestVM01.xva.tar.gz
Convert the image
For KVM machines we use raw or qcow2 images. I had to learn that xva images are in fact pretty weird tarballs with many single files that have to somehow be glued together.
Let's untar the image first:
tar xvf TestVM01.xva
You now have a folder with a name like "Ref:1912". In it are said files we have to somehow convert into a KVM image. After some Googleing I found a python script that could do that for me:
wget https://raw.githubusercontent.com/hswayne77/xenserver_to_xen/master/xenmigrate_new.py python ./xenmigrate_new.py -c Ref\:1912/ TestVM01.img
To be honest, I have no idea how the script works. There are a some similar scripts out there that claim to do the same – of those I tried, only this worked for me.
Prepare the image
Access the image
Xen doesn't need Grub as bootloader, so we have to install it. To do so, we have to be able to access the file system, so we have to mount the vm's system partition first and wire up basic system folders so we can chroot into it:
losetup /dev/loop0 TestVM01.img kpartx -v -a /dev/loop0 # maybe you have to install that one (apt-get install kpartx) losetup /dev/loop1 /dev/mapper/loop0p1 mkdir -p /mnt/vm mount /dev/loop1 /mnt/vm mount --bind /dev /mnt/vm/dev mount --bind /sys /mnt/vm/sys mount --bind /proc /mnt/vm/proc
Optional: In order to download and install grub, I first had to edit the settings to fit the new infrastructure:
/etc/resolve.conf– if it contains a DNS-server which is not available anymore, you can use 126.96.36.199 as a quick fix
Dealing with Grub is a bit tricky and you might need to play (and google) around with these settings. At first you need to install Grub2 and install it to the disk
aptitude update aptitude install grub2
Than set the
cat > /boot/grub/device.map <<EOF (hd0) /dev/loop0 (hd0,1) /dev/loop1 EOF
Finally I had to remove a setting, that attaches the vm to some kind of Xen terminal – since this environment won't be there anymore, this setting would prevent the VM from booting:
- Remove every occurrence of
- Remove the line(s) containing
I've also read that you have to check
/etc/fstab to point at the correct device. But in my case it referenced the partitions' UUIDs (and not a path like
/dev/xvda1), so I hadn't to do anything.
Install and update Grub:
grub-install /dev/loop0 update-grub
Check /etc/network/interfaces to fit the new infrastructure's requirements. We switched from fixed ip addresses to DHCP, so I had to change it like this:
# vi /etc/network/interfaces allow-hotplug eth0 iface eth0 inet dhcp
Leave the chroot:
Unwire all the mounts:
umount /mnt/vm/sys umount /mnt/vm/proc umount /mnt/vm/dev umount /mnt/vm losetup -d /dev/loop1 kpartx -v -d /dev/loop0 losetup -d /dev/loop0
Install the new VM
Before you setup the new VM, move the image to the path you want it to store permanently. Then install it using:
virt-install --name TestVM01 --ram 512 -f /path/to/TestVM01.img --import --vnc --connect qemu:///system
You now need to connect to the vm as quickly as you can via VNC (get the port using
virsh vncdisplay TestVM01). In my cases I had to press
e in the bootloader and remove the lines containing
loop1 in order to get Debian booting. Did the VM boot successfully you need to reinstall grub again:
grub-install /dev/vda update-grub
That did the trick for me! If you have any thoughts or tipps on this, please let me know!