Kata Containers with Cilium

Kata Containers is an open source project that provides a secure container runtime with lightweight virtual machines that feel and perform like containers, but provide stronger workload isolation using hardware virtualization technology as a second layer of defense. Kata Containers implements OCI runtime spec, just like runc that is used by Docker. Cilium can be used along with Kata Containers, using both enables higher degree of security. Kata Containers enhances security in the compute layer, while Cilium provides policy and observability in the networking layer.

Warning

Due to the different Kata Containers Networking model, there are limitations that can cause connectivity disruptions in Cilium. Please refer to the below Limitations section.

This guide shows how to install Cilium along with Kata Containers. It assumes that you have already followed the official Kata Containers installation user guide to get the Kata Containers runtime up and running on your platform of choice but that you haven’t yet setup Kubernetes.

Note

This guide has been validated by following the Kata Containers guide for Google Compute Engine (GCE) and using Ubuntu 18.04 LTS with the packaged version of Kata Containers, CRI-containerd and Kubernetes 1.18.3.

Setup Kubernetes with CRI

Kata Containers runtime is an OCI compatible runtime and cannot directly interact with the CRI API level. For this reason, it relies on a CRI implementation to translate CRI into OCI. At the time of writing this guide, there are two supported ways called CRI-O and CRI-containerd. It is up to you to choose the one that you want, but you have to pick one.

Refer to the section Requirements for detailed instruction on how to prepare your Kubernetes environment and make sure to use Kubernetes >= 1.12. Then, follow the official guide to run Kata Containers with Kubernetes.

Note

Minimum version of kubernetes 1.12 is required to use the RuntimeClass Feature for Kata Container runtime described below.

With your Kubernetes cluster ready, you can now proceed to deploy Cilium.

Deploy Cilium

Download the Cilium release tarball and change to the kubernetes install directory:

curl -LO https://github.com/cilium/cilium/archive/main.tar.gz
tar xzf main.tar.gz
cd cilium-main/install/kubernetes

Deploy Cilium release via Helm:

Using CRI-OUsing CRI-containerd

helm install cilium ./cilium \
  --namespace kube-system \
  --set bpf.autoMount.enabled=false

Warning

When using kube-proxy-replacement or its socket-level loadbalancer with Kata containers, the socket-level loadbalancer should be disabled for pods by setting socketLB.hostNamespaceOnly=true. See Socket LoadBalancer Bypass in Pod Namespace for more details.

Validate the Installation

Cilium CLIManually

Install the latest version of the Cilium CLI. The Cilium CLI can be used to install Cilium, inspect the state of a Cilium installation, and enable/disable various features (e.g. clustermesh, Hubble).

LinuxmacOSOther

CILIUM_CLI_VERSION=$(curl -s https://raw.githubusercontent.com/cilium/cilium-cli/main/stable.txt)
CLI_ARCH=amd64
if [ "$(uname -m)" = "aarch64" ]; then CLI_ARCH=arm64; fi
curl -L --fail --remote-name-all https://github.com/cilium/cilium-cli/releases/download/${CILIUM_CLI_VERSION}/cilium-linux-${CLI_ARCH}.tar.gz{,.sha256sum}
sha256sum --check cilium-linux-${CLI_ARCH}.tar.gz.sha256sum
sudo tar xzvfC cilium-linux-${CLI_ARCH}.tar.gz /usr/local/bin
rm cilium-linux-${CLI_ARCH}.tar.gz{,.sha256sum}

Clone the Cilium GitHub repository so that the Cilium CLI can access the latest unreleased Helm chart from the main branch:

git clone git@github.com:cilium/cilium.git
cd cilium

To validate that Cilium has been properly installed, you can run

$ cilium status --wait
   /¯¯\
/¯¯\__/¯¯\    Cilium:         OK
\__/¯¯\__/    Operator:       OK
/¯¯\__/¯¯\    Hubble:         disabled
\__/¯¯\__/    ClusterMesh:    disabled
   \__/

DaemonSet         cilium             Desired: 2, Ready: 2/2, Available: 2/2
Deployment        cilium-operator    Desired: 2, Ready: 2/2, Available: 2/2
Containers:       cilium-operator    Running: 2
                  cilium             Running: 2
Image versions    cilium             quay.io/cilium/cilium:v1.9.5: 2
                  cilium-operator    quay.io/cilium/operator-generic:v1.9.5: 2

Run the following command to validate that your cluster has proper network connectivity:

$ cilium connectivity test
ℹ️  Monitor aggregation detected, will skip some flow validation steps
✨ [k8s-cluster] Creating namespace for connectivity check...
(...)
---------------------------------------------------------------------------------------------------------------------
📋 Test Report
---------------------------------------------------------------------------------------------------------------------
✅ 69/69 tests successful (0 warnings)

Note

The connectivity test may fail to deploy due to too many open files in one or more of the pods. If you notice this error, you can increase the inotify resource limits on your host machine (see Pod errors due to “too many open files”).

Congratulations! You have a fully functional Kubernetes cluster with Cilium. 🎉

Run Kata Containers with Cilium CNI

Now that your Kubernetes cluster is configured with the Kata Containers runtime and Cilium as the CNI, you can run a sample workload by following these instructions.

Limitations

Due to its different Networking Design Architecture, the Kata runtime adds an additional layer of abstraction inside the Container Networking Namespace created by Cilium (referred to as “outer”). In that namespace, Kata creates an isolated VM with an additional Container Networking Namespace (referred to as “inside”) to host the requested Pod, as depicted below.

Upon the outer Container Networking Namespace creation, the Cilium CNI performs the following two actions:

1. creates the eth0 interface with the same device MTU of either the detected underlying network, or the MTU specified in the Cilium ConfigMap;

2. adjusts the default route MTU (computed as device MTU - overhead) to account for the additional networking overhead given by the Cilium configuration (ex. +50B for VXLAN, +80B for WireGuard, etc.).

However, during the inner Container Networking Namespace creation (i.e., the pod inside the VM), only the outer eth0 device MTU (1) is copied over by Kata to the inner eth0, while the default route MTU (2) is ignored. For this reason, depending on the types of connections, users might experience performance degradation or even packet drops between traditional pods and KataPod connections due to multiple (unexpected) fragmentation.

There are currently two possible workarounds, with (b) being preferred:

1. set a lower MTU value in the Cilium ConfigMap to account for the overhead. This would allow the KataPod to have a lower device MTU and prevent unwanted fragmentation. However, this is not recommended as it would have a relevant impact on all the other types of communications (ex. traditional pod-to-pod, pod-to-node, etc.) due to the lower device MTU value being set on all the Cilium-managed interfaces.

2. modify the KataPod deployment by adding an initContainer (with NET_ADMIN) to adjust the route MTU inside the inner pod. This would not only align the KataPod configuration to all the other pods, but also it would not harm all the other types of connections, given that it is a self-contained solution in the KataPod itself. The correct route MTU value to set can be either manually computed or retrieved by issuing ip route on a Cilium Pod (or inside a traditional pod). Here follows an example of a KataPod deployment (runtimeClassName: kata-clh) on a cluster with only Cilium VXLAN enabled (route MTU = 1500B - 50B = 1450):

   apiVersion: v1
   kind: Pod
   metadata:
     name: nginx-pod
     labels:
       app: nginx
   spec:
     runtimeClassName: kata-clh
     containers:
       - name: nginx
         image: nginx:latest
         ports:
           - containerPort: 80
     initContainers:
       - name: set-mtu
         image: busybox:latest
         command:
           - sh
           - -c
           - |
             DEFAULT="$(ip route show default)"
             ip route replace "$DEFAULT" mtu 1450
         securityContext:
           capabilities:
             add:
               - NET_ADMIN