Vagrant KVM Provider

This is a Vagrant 1.4+ plugin that adds a KVM provider to Vagrant, allowing Vagrant to control and provision KVM/QEMU VM.

Requirements

This plugin requires QEMU 1.1+, it has only been tested on Fedora 18+, Debian Wheezy, Ubuntu 12.04(LTS) Precise and Ubuntu 13.04 Raring at the moment.

This plugin requires several library and helper utils packages. Consult the Requirements section in INSTALL.md

Recent changes

Default image format is now qcow2 instead of sparsed raw image, with qcow2 vagrant-kvm uses the box volume as a backing volume so that VM creation is a lot faster. In most cases you want to use qcow2.

OVF boxes conversion has been removed, you should use vagrant-mutate instead.

Synced folders are now provided by a QEMU/KVM Virtfs in default. You can also use NFS for file share using type: "nfs" option.

There was a known libvirt bug in Ubuntu host: https://bugs.launchpad.net/ubuntu/+source/libvirt/+bug/943680 It was solved in Ubuntu 14.04(Trusty) and a backported libvirt provided by PPA above.

Features/Limitations

• Provides the same workflow as the Vagrant VirtualBox provider.
• Uses Virtfs for sync folders
• Only works with 1 VM per Vagrantfile for now
• Requires "libvirtd" group membership to run Vagrant (Debian/Ubuntu only)
• Requires backporting qemu and libvirt from experimental (Debian) or trusty (Ubuntu)
• Use qcow2 backing image by default, which should make VM creation very fast

Known issues

Vagrant 1.5 migration

Vagrant 1.5 changes a structure of user boxes directories. vagrant-KVM handle box directory as libvirt/qemu temporary spool, but Vagrant 1.5 changes it at first time launched.

Unfortunately vagrant-KVM 0.1.5 does not run on Vagrant-1.5.x. Users who interested in vagrant-kvm 0.1.5 may use with Vagrant 1.4.x.

This caused problem when following sinario:

1. user use vagrant-kvm 0.1.5 with vagrant-1.4.x
2. user upgrade vagrant 1.5.x
3. user upgrade vagrant 0.1.5.1 and after

We use transient pool instead of persistent one in vagrant-kvm 0.1.5.x, 0.1.6 and after Pools defined by vagrant-kvm will be removed after system reboot.

We recommend to use following combinations.

• Vagrant 1.3.x or before, and Vagrant-KVM 0.1.4

• Vagrant 1.5.x or after, and Vagrant-KVM 0.1.5.1 or after

If you are joining test for vagrant-kvm or other reasons you use vagrant-kvm 0.1.5 with vagrant 1.4.x, you got Call to virStoragePoolCreate failed: cannot open path error, please follow the instructions below. Please take care when running commands as root.

1. Upgrade vagrant-kvm to vagrant-kvm 0.1.5.1 or after

2. Clear all storage definitions.

$ sudo ls /etc/libvirt/storage/vagrant*
$ sudo rm /etc/libvirt/storage/vagrant*
$ sudo ls /etc/libvirt/storage/vagrant*

alternative way:

• open virt-manager
• connect to localhost
• right click and open details
• click storage tab
• right click 'vagrant*' storage pool and delete it.

3. restart libvirt daemon

Ubuntu/Debian

$ sudo service libvirt-bin restart

Fedora/CentOS/SuSE/Arch

$ sudo systemctl restart libvirtd

Ubuntu

Some versions of Ubuntu kernel has a bug that will cause vagrant-kvm to fail with a permission error on vagrant up. It is a kernel bug with the AppArmor security framework. As a workaround, please run following command to disable access control for libvirt helper.

sudo aa-complain /usr/lib/libvirt/virt-aa-helper

Fedora and Arch

With Fedora and Arch, default home directory permissions are set to drwx------. Qemu/kvm runs as 'qemu' user by default and cannot access your home directory.

To avoid this, please check and change your home directory and child directories permission to permit qemu user access to ~/.vagrant.d/tmp/storage-pool/

$ setfacl -m g:qemu:x /home/<your account>

Another option is to run qemu/kvm as the root user by changing the configuration in libvirt.

/etc/libvirt/qemu.conf

user = "root"
group = "root"

Then restart libvirtd.

$ sudo systemctl restart libvirtd

Another option is to specify an existing local storage pool with the storage-pool option for the KVM provider in your Vagrantfile (see below).

Usage

Install using the standard Vagrant 1.1+ plugin installation command. After installation, use vagrant up and specify the kvm provider. An example is shown below.

$ vagrant plugin install vagrant-kvm
$ vagrant up --provider=kvm

Of course prior to doing this, you'll need to obtain a KVM-compatible box file for Vagrant. You can convert a VirtualBox base box using vagrant-mutate https://github.com/sciurus/vagrant-mutate or see the sample box.

You will need a private network specifying an IP address in your Vagrantfile, the minimum Vagrantfile would then be:

Vagrant.configure("2") do |config|
  config.vm.box = "precise32"
  config.vm.network :private_network, ip: "192.168.192.10"
end

And then run vagrant up --provider=kvm.

If you always use kvm provider as default, please set it in your .bashrc:

export VAGRANT_DEFAULT_PROVIDER=kvm

then you can simply run vagrant up to use the kvm provider.

Configuration

There are some provider specific parameter to control VM definition.

• cpu_model - cpu architecture: 'i686' or 'x86_64': default is x86_64. Note that your base box should specify this.
• core_number - number of cpu cores.
• memory_size - memory size such as 512m, 1GiB, 100000KiB, unit is KiB if unspecified.
• storage_pool - specify an existing local storage pool to use instead of vagrant's own.
• gui - boolean for starting VM with VNC enabled.
• vnc_port - The port the VNC server listens to. Default is automatic port assignment.
• vnc_autoport - if true, KVM will automatically assign a port for VNC to listen to. Defaults to false, but the default vnc_port is -1, which results in this flag being automatically turned on by KVM.
• vnc_password - A password used to protect the VNC session.
• image_type - an image format for vm disk: 'raw' or 'qcow2': default is "qcow2" When choosing 'raw', vagrant-kvm always convert box image into storage-pool, which consumes disk space and is a longer process. Recommendation is 'qcow2'.
• machine_type - The type of machine to boot. Default is pc-1.2.
• network_model - The model of the network adapter you want to use. Defaults to virtio. Can be set to :default if you want to use the KVM default setting. Possible values include: ne2k_isa i82551 i82557b i82559er ne2k_pci pcnet rtl8139 e1000 virtio.
• video_model - The model of the video adapter. Default to cirrus. Can also be set to vga.
• image_mode - Possible value are clone or cow, defaults to cow. If set to clone, the image disk will be copied rather than use the original box image. This is slower but allows multiple VMs to be booted at the same time.
• customize - Customize virtual machine with virsh command. Similar functionality with virtualbox provider.
• disk_bus - disk interface to show virtual disk to guest: 'virtio' or 'sata', 'scsi' A box, which is 'mutate'-ed from virtualbox/vmware box, may specify sata/ide for disk bus. It may be useful to specify 'virtio' for performance, even when box defaults disk bus as sata/ide/scsi.
• seclabel - enables security labelling using selinux, apparmor, dac... based on the host distribution if set to on
• force_pause - use pause for vagrant suspend instead of suspend. It keeps resource online but execution is stopped. When VM has a device that is not supported hibernate, automatically use pause regardless of this.
• virtio_rng - boolean for optional virtio device of random number generator. QEMU 1.3.0 and after support this device. default false

Comparison with Vagrant-libvirt

Vagrant-kvm is a KVM/Qemu provider for single local host. Vagrant-kvm is simple, local host only, qemu/kvm only provider that is intended as an alternative to VirtualBox while keeping the same workflow.

Vagrant-libvirt is a libvirt provider to control machines via the libvirt toolkit. Vagrant-libvirt covers a lot more libvirt options, local and remote hosts and multiple hypervisors, such as Xen, LXC and KVM/qemu.

In early 2014, Vagrant-libvirt only supports kvm/qemu in local host, there is no big feature difference.

Here are a few difference:

1. Travis CI and quality assurance

Vagrant-kvm tests every pull request and repository with Travis-CI; https://travis-ci.org/adrahon/vagrant-kvm/ Vagrant-libvirt does not.

2. Copy-on-write

Vagrant-kvm by default uses the qcow2 format and qcow2 disk image backing to create a new volume from the box disk. This makes creating a new vm very fast, with some performance penalties. You can change this through configuration to clone(copy) disk image to pool instead. This will make new vm creation slower, but give you better disk performance. You can also configure vagrant-kvm to use raw images for an additional performance gain.

Vagrant-libvirt use qcow2 as disk image.

3. VNC port/password

Vagrant-kvm allows you to configure how to connect with VNC, which provides virtual guest desktop. Vagrant-libvirt does not.

4. Synced folder

Vagrant-kvm can provide synced folder with NFS only. We plan to provide virtfs(p9share) to allow sharing local folders without root privilege.

Vagrant-libvirt provide synced folder with Rsync and NFS. They also plan to support virtfs in future.

It is neccesary to fix several bugs in libvirt/qemu to enable the virtfs feature in both providers.

5. Snapshots via sahara

Vagrant-kvm plans to support snapshot via sahara. We have already proposed to sahara project to add support and are waiting for review. jedi4ever/sahara#32

Vagrant-libvirt is supported by sahara.

6. Use boxes from other Vagrant providers via vagrant-mutate

Both are supported by vagrant-mutate as convert target

7. Architecture

Vagrant-kvm controls kvm/qemu via ruby-libvirt, libvirt and qemu.

Vagrant-libvirt controls machines via fog, a cloud abstraction library in ruby, that is also used by vagrant-aws. The fog library controls virtual machines on supported platforms and provides control of qemu/kvm machines through ruby-libvirt and libvirt.