Kata Containers

Beginning with Charmed Kubernetes 1.16, the Kata Containers runtime can be used with containerd to safely run insecure or untrusted pods. When enabled, Kata provides hypervisor isolation for pods that request it, while trusted pods can continue to run on a shared kernel via runc. The instructions below demonstrate how to configure and use Kata containers.

Requirements

Due to their reliance on the KVM kernel module, Kata Containers can only be used on hosts that support virtualisation. Attempting to use Kata on a host that doesn't support virtualisation may result in an error similar to this one:

Failed create pod sandbox: rpc error: code = Unknown desc = failed to start sandbox container: failed to create containerd task: Could not access KVM kernel module: No such file or directory
qemu-vanilla-system-x86_64: failed to initialize KVM: No such file or directory

To fulfill this requirement, using Kata Containers on a public cloud may require a special instance type be used for the worker nodes. See the "Deploying to AWS" section below for an example.

Deploying Kata Containers

Kata Containers can be deployed to any Charmed Kubernetes cluster that's running with containerd.

Deploying with a new cluster

After bootstrapping a Juju controller, you can deploy Kata with Charmed Kubernetes by using the following YAML overlay:

applications:
  kata:
    charm: kata
relations:
- - kata:untrusted
  - containerd:untrusted
- - kata
  - kubernetes-control-plane
- - kata
  - kubernetes-worker

Save this YAML and then deploy:

juju deploy charmed-kubernetes --overlay kata.yaml

Deploying to an existing cluster

Deploy the Kata charm and add the necessary relations using the following commands:

juju deploy kata
juju integrate kata kubernetes-control-plane
juju integrate kata kubernetes-worker
juju integrate kata:untrusted containerd:untrusted

Deploying to AWS

A convenient way to try Charmed Kubernetes with Kata Containers is to deploy it to AWS. A special AWS instance type is required to provide the necessary virtualisation support. To deploy Charmed Kubernetes with Kata Containers on AWS, write the following overlay to a file or download it here:

applications:
  kubernetes-worker:
    constraints: instance-type=i3.metal
    num_units: 1
  kata:
    charm: kata
relations:
- - kata:untrusted
  - containerd:untrusted
- - kata
  - kubernetes-control-plane
- - kata
  - kubernetes-worker

Once written, deploy it with:

juju deploy charmed-kubernetes --overlay kata-aws-overlay.yaml

Due to the high costs of i3.metal instance types, the example above deploys only one worker node. Feel free to edit the num_units field to suit your needs, or add more workers later with the juju add-unit kubernetes-worker command.

Deploying pods to Kata

Untrusted annotation

The simplest way to run your pods with Kata is to annotate them with io.kubernetes.cri.untrusted-workload: "true". For example:

apiVersion: v1
kind: Pod
metadata:
  name: nginx-untrusted
  annotations:
    io.kubernetes.cri.untrusted-workload: "true"
spec:
  containers:
  - name: nginx
    image: nginx

RuntimeClass

If you don't want to taint your workloads as untrusted, you can also create the following RuntimeClass:

apiVersion: node.k8s.io/v1beta1
kind: RuntimeClass
metadata:
  name: kata
handler: kata

After this RuntimeClass is created, workloads can be run which are pinned to the class.

apiVersion: v1
kind: Pod
metadata:
  name: nginx-kata
spec:
  runtimeClassName: kata
  containers:
  - name: nginx
    image: nginx