Salt as a Cloud Controller

In Salt 0.14.0, an advanced cloud control system was introduced, allowing private cloud VMs to be managed directly with Salt. This system is generally referred to as Salt Virt.

The Salt Virt system already exists and is installed within Salt itself. This means that besides setting up Salt, no additional salt code needs to be deployed.

Note

The libvirt python module and the certtool binary are required.

The main goal of Salt Virt is to facilitate a very fast and simple cloud that can scale and is fully featured. Salt Virt comes with the ability to set up and manage complex virtual machine networking, powerful image and disk management, and virtual machine migration with and without shared storage.

This means that Salt Virt can be used to create a cloud from a blade center and a SAN, but can also create a cloud out of a swarm of Linux Desktops without a single shared storage system. Salt Virt can make clouds from truly commodity hardware, but can also stand up the power of specialized hardware as well.

Setting up Hypervisors

The first step to set up the hypervisors involves getting the correct software installed and setting up the hypervisor network interfaces.

Installing Hypervisor Software

Salt Virt is made to be hypervisor agnostic but currently, the only fully implemented hypervisor is KVM via libvirt.

The required software for a hypervisor is libvirt and kvm. For advanced features, install libguestfs or qemu-nbd.

Note

Libguestfs and qemu-nbd allow for virtual machine images to be mounted before startup and get pre-seeded with configurations and a salt minion.

This sls will set up the needed software for a hypervisor, and run the routines to set up the libvirt pki keys.

Note

Package names and setup used is Red Hat specific. Different package names will be required for different platforms.

libvirt:
  pkg.installed: []
  file.managed:
    - name: /etc/sysconfig/libvirtd
    - contents: 'LIBVIRTD_ARGS="--listen"'
    - require:
      - pkg: libvirt
  virt.keys:
    - require:
      - pkg: libvirt
  service.running:
    - name: libvirtd
    - require:
      - pkg: libvirt
      - network: br0
      - libvirt: libvirt
    - watch:
      - file: libvirt

libvirt-python:
  pkg.installed: []

libguestfs:
  pkg.installed:
    - pkgs:
      - libguestfs
      - libguestfs-tools

Hypervisor Network Setup

The hypervisors will need to be running a network bridge to serve up network devices for virtual machines. This formula will set up a standard bridge on a hypervisor connecting the bridge to eth0:

eth0:
  network.managed:
    - enabled: True
    - type: eth
    - bridge: br0

br0:
  network.managed:
    - enabled: True
    - type: bridge
    - proto: dhcp
    - require:
      - network: eth0

Virtual Machine Network Setup

Salt Virt comes with a system to model the network interfaces used by the deployed virtual machines. By default, a single interface is created for the deployed virtual machine and is bridged to br0. To get going with the default networking setup, ensure that the bridge interface named br0 exists on the hypervisor and is bridged to an active network device.

Note

To use more advanced networking in Salt Virt, read the Salt Virt Networking document:

Salt Virt Networking

Libvirt State

One of the challenges of deploying a libvirt based cloud is the distribution of libvirt certificates. These certificates allow for virtual machine migration. Salt comes with a system used to auto deploy these certificates. Salt manages the signing authority key and generates keys for libvirt clients on the master, signs them with the certificate authority, and uses pillar to distribute them. This is managed via the libvirt state. Simply execute this formula on the minion to ensure that the certificate is in place and up to date:

Note

The above formula includes the calls needed to set up libvirt keys.

libvirt_keys:
  virt.keys

Getting Virtual Machine Images Ready

Salt Virt requires that virtual machine images be provided as these are not generated on the fly. Generating these virtual machine images differs greatly based on the underlying platform.

Virtual machine images can be manually created using KVM and running through the installer, but this process is not recommended since it is very manual and prone to errors.

Virtual Machine generation applications are available for many platforms:

kiwi: (openSUSE, SLES, RHEL, CentOS)

https://opensuse.github.io/kiwi/

vm-builder:

https://wiki.debian.org/VMBuilder

Once virtual machine images are available, the easiest way to make them available to Salt Virt is to place them in the Salt file server. Just copy an image into /srv/salt and it can now be used by Salt Virt.

For purposes of this demo, the file name centos.img will be used.

Existing Virtual Machine Images

Many existing Linux distributions distribute virtual machine images which can be used with Salt Virt. Please be advised that NONE OF THESE IMAGES ARE SUPPORTED BY SALTSTACK.

CentOS

These images have been prepared for OpenNebula but should work without issue with Salt Virt, only the raw qcow image file is needed: https://wiki.centos.org/Cloud/OpenNebula

Fedora Linux

Images for Fedora Linux can be found here: https://alt.fedoraproject.org/cloud

Ubuntu Linux

Images for Ubuntu Linux can be found here: http://cloud-images.ubuntu.com/

Using Salt Virt

With hypervisors set up and virtual machine images ready, Salt can start issuing cloud commands using the virt runner.

Start by running a Salt Virt hypervisor info command:

salt-run virt.host_info

This will query the running hypervisor(s) for stats and display useful information such as the number of CPUs and amount of memory.

You can also list all VMs and their current states on all hypervisor nodes:

salt-run virt.list

Now that hypervisors are available a virtual machine can be provisioned, the virt.init routine will create a new virtual machine:

salt-run virt.init centos1 2 512 salt://centos.img

The Salt Virt runner will now automatically select a hypervisor to deploy the new virtual machine on. Using salt:// assumes that the CentOS virtual machine image is located in the root of the Salt File Server on the master. When images are cloned (i.e. copied locally after retrieval from the file server), the destination directory on the hypervisor minion is determined by the virt:images config option; by default this is /srv/salt-images/.

When a VM is initialized using virt.init, the image is copied to the hypervisor using cp.cache_file and will be mounted and seeded with a minion. Seeding includes setting pre-authenticated keys on the new machine. A minion will only be installed if one can not be found on the image using the default arguments to seed.apply.

Note

The biggest bottleneck in starting VMs is when the Salt Minion needs to be installed. Making sure that the source VM images already have Salt installed will GREATLY speed up virtual machine deployment.

You can also deploy an image on a particular minion by directly calling the virt execution module with an absolute image path. This can be quite handy for testing:

salt 'hypervisor*' virt.init centos1 2 512 image=/var/lib/libvirt/images/centos.img

Now that the new VM has been prepared, it can be seen via the virt.query command:

salt-run virt.query

This command will return data about all of the hypervisors and respective virtual machines.

Now that the new VM is booted, it should have contacted the Salt Master. A test.ping will reveal if the new VM is running.

QEMU Copy on Write Support

For fast image cloning, you can use the qcow disk image format. Pass the enable_qcow flag and a .qcow2 image path to virt.init:

salt 'hypervisor*' virt.init centos1 2 512 image=/var/lib/libvirt/images/centos.qcow2 enable_qcow=True start=False

Note

Beware that attempting to boot a qcow image too quickly after cloning can result in a race condition where libvirt may try to boot the machine before image seeding has completed. For that reason, it is recommended to also pass start=False to virt.init.

Also know that you must not modify the original base image without first making a copy and then rebasing all overlay images onto it. See the qemu-img rebase usage docs.

Migrating Virtual Machines

Salt Virt comes with full support for virtual machine migration. Using the libvirt state in the above formula makes migration possible.

A few things need to be available to support migration. Many operating systems turn on firewalls when originally set up; the firewall needs to be opened up to allow for libvirt and kvm to cross communicate and execution migration routines. On Red Hat based hypervisors in particular, port 16514 needs to be opened on hypervisors:

iptables -A INPUT -m state --state NEW -m tcp -p tcp --dport 16514 -j ACCEPT

Note

More in-depth information regarding distribution specific firewall settings can be found in:

Opening the Firewall up for Salt

Salt also needs the virt:tunnel option to be turned on. This flag tells Salt to run migrations securely via the libvirt TLS tunnel and to use port 16514. Without virt:tunnel, libvirt tries to bind to random ports when running migrations.

To turn on virt:tunnel, simply apply it to the master config file:

virt:
    tunnel: True

Once the master config has been updated, restart the master and send out a call to the minions to refresh the pillar to pick up on the change:

salt \* saltutil.refresh_modules

Now, migration routines can be run! To migrate a VM, simply run the Salt Virt migrate routine:

salt-run virt.migrate centos <new hypervisor>

VNC Consoles

Although not enabled by default, Salt Virt can also set up VNC consoles allowing for remote visual consoles to be opened up. When creating a new VM using virt.init, pass the enable_vnc=True parameter to have a console configured for the new VM.

The information from a virt.query routine will display the VNC console port for the specific VMs:

centos
  CPU: 2
  Memory: 524288
  State: running
  Graphics: vnc - hyper6:5900
  Disk - vda:
    Size: 2.0G
    File: /srv/salt-images/ubuntu2/system.qcow2
    File Format: qcow2
  Nic - ac:de:48:98:08:77:
    Source: br0
    Type: bridge

The line Graphics: vnc - hyper6:5900 holds the key. First the port named, in this case 5900, will need to be available in the hypervisor's firewall. Once the port is open, then the console can be easily opened via vncviewer:

vncviewer hyper6:5900

By default there is no VNC security set up on these ports, which suggests that keeping them firewalled and mandating that SSH tunnels be used to access these VNC interfaces. Keep in mind that activity on a VNC interface that is accessed can be viewed by any other user that accesses that same VNC interface, and any other user logging in can also operate with the logged in user on the virtual machine.

Conclusion

Now with Salt Virt running, new hypervisors can be seamlessly added just by running the above states on new bare metal machines, and these machines will be instantly available to Salt Virt.