Deploying a KVM (x86_64) Device

Adding a KVM device to LAVA is an easy way to make sure things work without having to worry about connecting to a physical device and setting up a master image. This page outlines the steps required to add a new KVM device to your LAVA deployment and make it able to accept job requests.

Obtain an image

A variety of pre-built images are available for download:

http://images.validation.linaro.org/kvm/

Building KVM images for LAVA

If the prebuilt images are not suitable, new images can be built using vmdebootstrap and a couple of custom overlay packages. However, image building tools are constantly developing and this guide can easily become out of date, so you will need to read the manpages for the tools and update the instructions appropiately.

The overlays are only necessary if you are not going to use the automated login support in LAVA. See Deploying an image to a device for the current dispatcher and Writing a job submission in YAML for the pipeline dispatcher.

git clone git://git.linaro.org/lava/lava-vmdebootstrap.git

lava-vmdebootstrap is just a small wrapper around vmdebootstrap which does all the hard work. What lava-vmdebootstrap does is download the Linaro image overlays and pass some options that we will always need so that you don’t need to.

The default distribution is Debian stable. Add the ``–distribution=wheezy` option to change to wheezy.

If you choose to use the installed vmdebootstrap package, tweak the file:lava-vmdebootstrap call to change ./vmdebootstrap to vmdebootstrap.

To build wheezy images on Debian stable, you may need to install the mbr package as well.

Example invocation:

$ sudo ./lava-vmdebootstrap --image=myimage.img

To run the test image, make sure it is writeable:

$ sudo chmod a+w ./myimage.img

(Newer versions of vmdebootstrap can do this for you if you pass the --owner=$(whoami) option.

Execute using qemu, e.g. on amd64 using qemu-system-x86_64:

$ qemu-system-x86_64 ./myimage.img

See man 1 vmdebootstrap and man qemu-system for more information.

Once the overlay packages have been downloaded, you can call vmdebootstrap directly to create other types of images without needing to modify lava-vmdebootstrap. e.g. this is a call to vmdebootstrap to create a KVM image based on Ubuntu:

sudo vmdebootstrap \
 --custom-package='linaro-overlay_1112.2_all.deb' \
 --custom-package='linaro-overlay-minimal_1112.2_all.deb' \
 --enable-dhcp --no-kernel --package=linux-image \
 --serial-console --serial-console-command='/bin/auto-serial-console' \
 --root-password='root' --hostname='ubuntu' --user=linaro/linaro --sudo \
 --verbose --image=myimage.img

This command extends lava-vmdebootstrap to make a 4G LAVA KVM image based on Debian testing using the UK Debian mirror:

sudo vmdebootstrap \
 --custom-package='linaro-overlay_1112.2_all.deb' \
 --custom-package='linaro-overlay-minimal_1112.2_all.deb' \
 --enable-dhcp \
 --serial-console --serial-console-command='/bin/auto-serial-console' \
 --root-password='root' \
 --distribution testing --size 4g \
 --mirror http://ftp.uk.debian.org/debian \
 --verbose --image=myimage.img

Adding a KVM device to LAVA

You can use the Instance name support:

$ sudo /usr/share/lava-server/add_device.py kvm kvm01

Configure the dispatcher manually

Create your kvm01.conf file with the following content:

device_type = kvm
root_part = 1

Sample job file (replace file:///path/to/kvm.img with the actual location where you placed the image you created in the previous step):

{
  "timeout": 18000,
  "job_name": "kvm-test",
  "device_type": "kvm",
  "target": "kvm01",
  "actions": [
    {
      "command": "deploy_linaro_image",
      "parameters": {
        "image": "file:///path/to/kvm.img"
        }
    },
    {
      "command": "boot_linaro_image"
    }
  ]
}

To test, you can execute the dispatcher directly with the following command as root:

lava-dispatch /tmp/kvm.json

Optional: networking configuration

By default, LAVA kvm devices will use virtio networking, which is a lot faster than the QEMU default at the time of writing this. But the default configuration also uses NAT, which makes the virtual machines unacessible from other hosts in your local network.

Setting up a TAP device for KVM networking is a way to both make networking faster and make the virtual machines available from other nodes in the network.

This requires some extra configuration, and that’s why it’s not the default. It goes like this:

Device configuration file(kvmXX.conf):

device_type = kvm
root_part = 1
kvm_networking_options = -net nic,model=virtio -net tap

Then add a bridge interface to the networking configuration (/etc/network/interfaces). Example:

auto eth0
iface eth0 inet manual

auto br0
    iface br0 inet dhcp
    bridge_ports eth0
    bridge_stp off
    bridge_fd 0
    bridge_maxwait 0

Please note the above are examples, as we do not want to duplicate the QEMU documentation. Make sure you consult the official QEMU documentation for detailed instructions on how to create a proper TAP interface setup.

Configuring the scheduler manually

Now that the dispatcher understand the KVM device and can work with it, we need to inform the LAVA scheduler about it. This is done from the admin panel in the LAVA web app.

You’ll first add a “kvm” device type by going to a URL like:

http://localhost/admin/lava_scheduler_app/devicetype/

That page will give you an option to add a device type. From the add device type page, you need to give the name “kvm”. Don’t touch any of the other options for now.

After adding a device type you can add a device. From this page you’ll want to set the hostname to the same value you set for ‘target’ in the dispatch config. Then select “kvm” from the device type list.

Now when you view:

http://localhost/scheduler/

You should see your new device type and be able to drill down to the device.

Submitting a KVM Job

The scheduler documentation includes instructions for Job Submission to LAVA. You can use the job file shown above as the basis for your new job.