Learn how to inject workloads with Kata Containers in Istio.


Kata Containers, which is a merger of two existing open source projects – Intel Clear Containers and Hyper runV, is an open source project and community working to build a standard implementation of lightweight Virtual Machines (VMs) that feel and perform like containers, but provide the workload isolation and security advantages of VMs.

Based on the definition above, it’s natural for me to think about how to achieve the Kata Containers as one of multiple container runtimes in Istio to inject the workload. The good news is that I successfully deployed it in my environment, and this blog will introduce what benefits we can get from Kata Containers and how to deploy it.

What’s the benefit of using Kata Containers?

Kata Containers is designed to provide greater isolation between containers while still maintaining the performance and efficiency provided by other runtimes. There is one critical difference for Kata Containers to be a special container runtime: Kata runtime enforces a deeper level of isolation between containers than other runtimes by following below two points:

  • Each pod runs its own kernel instead of sharing the host system’s kernel with the host and other containers using cgroups in Kata
  • Kata Containers can take advantage of security features provided by hardware-level virtualization (meaning virtualization that is built into CPUs and made available using VT extensions)

However, it’s different from common virtual machines to take a minute or two for starting and wasting a fair amount of hardware resources on establishing container isolation, Kata starts just as fast and consumes resources just as efficiently as other containers. The containers become more secure with the lowest cost:

How to deploy workload injection with Kata Containers in Istio:

Environment Prepared

  • Containerd 1.5.6
  • Kata Runtime v3.0.2
  • Kubernetes Cluster via kubeadm (do not use Kind) v1.23
  • Istio v1.17-dev

Install Kata Containers

There are six methods to install Kata Containers via the installation guides found here and here. I have tried two of them (Using snap and Manual), and I recommend the manual method because it’s a mess when using snap. So let’s get into Manual guides.

Key highlights:

1. After untarring the Kata Containers package, move all files into /opt/kata or /$HOME/opt/kata directory, depending on your system & user. Once complete, there should be four files — bin , libexecruntime-rs, and share under /opt/kata .

Copy /opt/kata/share/defaults/kata-containers/configuration.toml into /etc/kata-containers/configuration.toml Note: Please choose the corrected configuration file for you, and create the /etc/kata-containers directory if it does not exist.

2. Create five symbolic links (symlinks) for Kata Containers installation files by following           commands:

$ ln -s /opt/kata/bin/kata-runtime /usr/local/bin/kata-runtime
$ ln -s /opt/kata/bin/containerd-shim-kata-v2 /usr/local/bin/containerd-shim-kata-v2
$ ln -s /opt/kata/bin/kata-monitor /usr/local/bin/kata-monitor
$ ln -s /opt/kata/bin/kata-collect-data.sh /usr/local/bin/kata-collect-data.sh
$ ln -s /opt/kata/bin/qemu-system-x86_64 /usr/local/bin/qemu-system-x86_64

Please refer to Kata Containers installation guides for more detail.

Configure Containerd correctly

This content is included in Configuring Containerd to use Kata Containers, however, I just want to simplify the plugins configuration for Containerd by using RuntimeClass as shown below:

default_runtime_name = “runc”
runtime_type = “io.containerd.kata.v2”
privileged_without_host_devices = true
ConfigPath = “/opt/kata/share/defaults/kata-containers/configuration.toml”
runtime_type = “io.containerd.runc.v2”
SystemdCgroup = true

In this configuration, I define two types of runtime: runc and kata and default by runc.

Note: If Containerd configuation is changed, you will need to reload & restart services using the commands sudo systemctl daemon-reload and sudo systemctl restart containerd.

Configure kubelet correctly

This content is included in Configure kubelet to use Kata Containers, the configuration file /etc/systemd/system/kubelet.service.d/10-kubeadm.conf is below:

$ Environment=”KUBELET_EXTRA_ARGS= — container-runtime=remote — container-runtime-endpoint=unix:///run/containerd/containerd.sock — cgroup-driver=systemd”

Inject workloads with Kata Containers in Istio

Generally, there should be no difference when installing k8s cluster and Istio, demonstrated below as I install cluster via kubeadm command:

$ kubeadm init — cri-socket=unix:///run/containerd/containerd.sock — pod-network-cidr= — v=5 — ignore-preflight-errors=all`

Then install Istio via istioctl command istioctl install -y

Based on the Containerd configuration from before, runc is the default container runtime, which can be verified with the command: $ sudo systemctl status containerd

I use RuntimeClass to launch the Kata Containers, using the yaml file kata-runtimeclass.yaml below:

apiVersion: node.k8s.io/v1
kind: RuntimeClass
  name: kata-runtime
handler: kata

The RuntimeClass CustomResource (CR) name is kata-runtime and the handler field is specified as kata. Create this CR by command:

$ kubectl apply -f kata-runtimeclass.yaml

I use httpbin as an example by using httpbin.yaml, however, there should be a little change for it, and the final yaml is below:

apiVersion: v1
kind: ServiceAccount
 name: httpbin
- -
apiVersion: v1
kind: Service
 name: httpbin
  app: httpbin
  service: httpbin
 - name: http
 port: 8000
 targetPort: 80
 app: httpbin
 - -
apiVersion: apps/v1
kind: Deployment
 name: httpbin
 replicas: 1
   app: httpbin
   version: v1
    app: httpbin
    version: v1
   runtimeClassName: kata-runtime
   serviceAccountName: httpbin
   - image: docker.io/kennethreitz/httpbin
     imagePullPolicy: IfNotPresent
     name: httpbin
     - containerPort: 80

Be mindful of the presence of an extra runtimeClassName: kata-runtime field in the containers section of the Deployment. I have already defined the RuntimeClass CR as kata-runtime. Another important note is that the httpbin ServiceAccount is required, because Kata Containers needs a cluster role.

Now it is time to deploy this service using the following command:

$ kubectl apply -f <(istioctl kube-inject -f httpbin-hsm.yaml)

Verify the workload as demonstrated below:

There should be a new Runtime Class Name property, as is shown in the above red rectangle box.

Besides the Containerd containerd-shim-runc-v2, there are also other processes for Kata Containers shown in the above red rectangle box, such as containerd-shim-kata-v2virtiofsd and qemu-system-x86_64. This is because I deploy a service which requires Kata Containers to inject in Cluster.


  • Basically, there are not many differences for using Kata Containers with other generic container runtimes in Istio from my use case: deploy httpbin and bookinfo, but we still need to verify more other things, such as network CNI situations, Istio CNI enablement and so on.
  • There may be multiple container runtimes according to different requirements in Istio later in the same environment at the same time, and we need to be aware of this.

For more articles about Kata Containers, check out katacommunity.medium.com.

Want to write a blog post or have an idea for a Superuser article? Go to openinfrafoundation.formstack.com/forms/superuser_pitch and submit your pitch. 

Steve(Huailong) Zhang
Latest posts by Steve(Huailong) Zhang (see all)