charm-kube-virt

Charmhub package name: kubevirt More information: https://charmhub.io/kubevirt

As an optional addon to charmed-kubernetes, this charm allows for installation of a selectable version of kubevirt into a cluster. KubeVirt allows for virtual machine instances to be launched as kubernetes workloads which interact with other internal services within a kubernetes cluster.

Basics

• Requires a charmed-kubernetes deployment launched by juju
• Deploy the kubevirt charm in the model relating to the control-plane and to the worker
• Once the model is active/idle, the kubevirt charm will have successfully deployed KubeVirt into the cluster.
• Confirm the kubevirt object is deployed with kubectl get kubevirt -n kubevirt

Troubleshooting

• kubevirt requires that the control-plane allows privileged containers.

  applications:
  kubernetes-control-plane:
    options:
      allow-privileged: "true"

• If the worker machines are themselves VMs, you will be creating VMs in VMs or (nested virtualization)

  • the charm will enable software-emulation for virtualization if /dev/kvm is not detected on any one worker node.

Contributing

Please see the Juju SDK docs for guidelines on enhancements to this charm following best practice guidelines, and CONTRIBUTING.md for developer guidance.

Installation

Deploying the charm into an existing juju machine model with charmed-kubernetes is as simple as

juju deploy kubevirt --channel <channel> --config operator-release='<v0.min.bug>' --trust

Next, the charm will need to be related to every kubernetes-worker in the model. If you have multiple kubernetes-worker applications, you may relate the same charm to those multiple applications.

juju relate kubernetes-worker kubevirt:juju-info

Lastly, the charm will need to relate to the control-plane charm in the model.

juju relate kubernetes-control-plane kubevirt:kube-control

This will install a subordinate unit on every worker, use its cluster credentials to install the KubeVirt manifests into the cluster, and detect if the related workers all have access to an internal /dev/kvm device. The necessary binaries will be installed on the worker machines using apt (fundamentally qemu-system-* or qemu) and apparmor adjustments will be made after the installation.

Networking

Within this repository, we will be testing with the default charmed-kubernetes CNI (calico) and will not be testing with multiple network interfaces per VM instance. With a stock CNI, your VM will have access to a single network interface -- the same that is used to communicate with the rest of the cluster. Fortunately, not much precludes KubeVirt from accessing other network interfaces through more advanced CNIs.

See Usage for more details

Storage

Within this repository, we will be testing with a VMWare cluster which can provider storage through the vsphere-cloud-provider charm. This charm can allocate block storage in the VMWare cluster, mount the block storage into the VM, and use it for persistent storage through a PVC. The very same methodology may be used within other clouds with storage classes and PVCs.

Usage

Once installed, follow upstream documenation for complete usage guides.

What to expect:

• a kubevirt namespace will be created which contains
  • the replicaset resources: virt-api, virt-controller, and virt-operator
  • the daemonset resource: virt-handler
  • the KubeVirt resource: kubevirt

Any configuration changes to the KubeVirt resource should be considered properies of this charm though not all have yet been implemented.

• image-registry: per upstream, the kubevirt images are hosted on quay.io
• software-emulation: the charm will detect if all the workers have access to /dev/kvm.
  • If true, then configuration.developerConfiguration.useEmulation will be false
  • If one machine doesn't have /dev/kvm, then the charm is forced to enable useEmulation
• pvc-tolerate-less-space-up-to-percent: is an adjustment necessary under certain circumstances where the cloud provider's storage isn't quite sized to create filesystems with enough available space see upstream docs for more guidance.

ContainerDisk Images

ContainerDisks are a specialized volume type for Virtual Machine instances. They consist of a file (qcow2 or raw) in the /disk directory with permissions o107.

See upstream docs to create your own disk images.

Networking [NetworkingUsage]

Virtual Machines will have one pod interfaces, and can add multiple multus network interfaces provided via multus.

Multus can be installed into a kubernetes cluster with the multus charm

See the upstream docs to determine the VMI configuration for multiple nics.

Storage

Virtual Machine instances may be booted with multiple disk devices, each of which can be defined by various kubernetes storage components. See Supported Volume Types for a complete list of volume types.

VM Management

Creating and Destroying the VM instances is handled through the kubernetes api using kubectl commands, but some VM management takes place using the virt-api service in the kubernetes cluster. One may use the virtctl client directly or install a virtctl plugin into kubectl. See upstream docs for more details on using virtctl. For ease of use this charm can expose virtctl through juju run

juju run --unit kubevirt/leader -- ./virtctl help