Virtualization With KVM On Ubuntu 8.10 - Page 2
3 Creating An Image-Based VM
We can now create our first VM - an image-based VM (if you expect lots of traffic and many read- and write operations for that VM, use an LVM-based VM instead as shown in chapter 6 - image-based VMs are heavy on hard disk IO).
We will create a new directory for each VM that we want to create, e.g. ~/vm1, ~/vm2, ~/vm3, and so on, because each VM will have a subdirectory called ubuntu-kvm, and obviously there can be just one such directory in ~/vm1, for example. If you try to create a second VM in ~/vm1, for example, you will get an error message saying ubuntu-kvm already exists (unless you run vmbuilder with the --dest=DESTDIR argument):
root@server1:~/vm1# vmbuilder kvm ubuntu -c vm2.cfg
We will use the vmbuilder tool to create VMs. (You can learn more about vmbuilder here.) vmbuilder uses a template to create virtual machines - this template is located in the /etc/vmbuilder/libvirt/ directory. Because we must modify the template, we create a copy and modify that one:
mkdir -p ~/vm1/mytemplates/libvirt
Now we open ~/vm1/mytemplates/libvirt/libvirtxml.tmpl...
... and change the network section from
because we want the VM to use our network bridge.
Now we come to the partitioning of our VM. We create a file called vmbuilder.partition...
... and define the desired partitions as follows:
This defines a root partition (/) with a size of 8000MB, a swap partition of 4000MB, and a /var partition of 20000MB. The --- line makes that the following partition (/var in this example) is on a separate disk image (i.e., this would create two disk images, one for root and swap and one for /var). Of course, you are free to define whatever partitions you like (as long as you also define root and swap), and of course, they can be in just one disk image - this is just an example.
I want to install openssh-server in the VM. To make sure that each VM gets a unique OpenSSH key, we cannot install openssh-server when we create the VM. Therefore we create a script called boot.sh that will be executed when the VM is booted for the first time. It will install openssh-server (with a unique key) and also force the user (I will use the default username administrator for my VMs together with the default password howtoforge) to change the password when he logs in for the first time:
Make sure you replace the username administrator with your default login name.
(You can find more about this here: https://help.ubuntu.com/community/JeOSVMBuilder#First%20boot)
(You can also define a "first login" script as described here: https://help.ubuntu.com/community/JeOSVMBuilder#First%20login)
Whenever vmbuilder builds a new VM, it has to download all packages from an Ubuntu mirror which can take quite some time. To speed this up, we install apt-proxy...
apt-get install apt-proxy
... to cache the downloaded packages so that subsequent VM installations will be a lot faster.
Now open /etc/apt-proxy/apt-proxy-v2.conf...
... and replace the default Ubuntu mirror with a mirror close to you (e.g. http://de.archive.ubuntu.com/ubuntu if you are in Germany):
Then we restart apt-proxy:
apt-proxy listens on port 9999, so we can pass our local apt-proxy "mirror" as an argument to the vmbuilder script.
Now take a look at
vmbuilder kvm ubuntu --help
to learn about the available options.
To create our first VM, vm1, we go to the VM directory...
... and run vmbuilder, e.g. as follows:
vmbuilder kvm ubuntu --suite=intrepid --flavour=virtual --arch=amd64 --mirror=http://192.168.0.100:9999/ubuntu -o --libvirt=qemu:///system --tmpfs=- --ip=192.168.0.101 --part=vmbuilder.partition --templates=mytemplates --user=administrator --name=Administrator --pass=howtoforge --addpkg=vim-nox --addpkg=unattended-upgrades --addpkg=acpid --firstboot=boot.sh --mem=256 --hostname=vm1
Most of the options are self-explanatory. --part specifies the file with the partitioning details, relative to our working directory (that's why we had to go to our VM directory before running vmbuilder), --templates specifies the directory that holds the template file (again relative to our working directory), and --firstboot specifies the firstboot script. --libvirt=qemu:///system tells KVM to add this VM to the list of available virtual machines. --addpkg allows you to specify Ubuntu packages that you want to have installed during the VM creation (see above why you shouldn't add openssh-server to that list and use the firstboot script instead).
In the --mirror line I have specified my local apt-proxy mirror (http://192.168.0.100:9999/ubuntu) - I have used my publically accessible IP address instead of localhost or 127.0.0.1 because this mirror will be used in the VM's /etc/apt/sources.list file as well, and of course, the VM won't be able to connect to 127.0.0.1 on the host. Of course, you can as well specify an official Ubuntu repository in --mirror, e.g. http://de.archive.ubuntu.com/ubuntu. If you leave out --mirror, then the default Ubuntu repository (http://archive.ubuntu.com/ubuntu) will be used.
The build process can take a few minutes.
Afterwards, you can find an XML configuration file for the VM in /etc/libvirt/qemu/ (=> /etc/libvirt/qemu/vm1.xml):
ls -l /etc/libvirt/qemu/
root@server1:~/vm1# ls -l /etc/libvirt/qemu/
The disk images are located in the ubuntu-kvm/ subdirectory of our VM directory:
ls -l ~/vm1/ubuntu-kvm/
root@server1:~/vm1# ls -l ~/vm1/ubuntu-kvm/
4 Creating A Second VM
If you want to create a second VM (vm2), here's a short summary of the commands:
mkdir -p ~/vm2/mytemplates/libvirt
(Please note that you don't have to create a new directory for the VM (~/vm2) if you pass the --dest=DESTDIR argument to the vmbuilder command - it allows you to create a VM in a directory where you've already created another VM. In that case you don't have to create new vmbuilder.partition and boot.sh files and don't have to modify the template, but can simply use the existing files: