A Calustra - Eloy Coto Pereiro Home About Books

Terraform recipes to test Cilium on Kubernetes


After almost five years, I left my job at Foehn; I need a personal break, and I have decided early this year that late August would be a good time for doing it.

During this break, I'll be doing open source for few months before starting to look for a new place. I want to learn from the best software engineers, so I'll be contributing the next two months to Cilium project.

If you don't know what is Cilium let me explain, Cilium is a software that enables network security mainly on container related projects, like Docker, Kuberntes or Itsio. The main difference between Cilium and other networking solutions is that Cilium relies in a heavy way on a lot of new technology that the kernel provides.


Firstly, we need to understand what BPF (Berkeley Packet Filter) means. BPF is a Linux kernel bytecode interpreter originally introduced to filter network packets (tcpdump, ngrep).

A BPF program is compiled to bytecode by the JIT compiler, it becomes a native CPU execution, as a consequence of that, the filter performance increases a lot.

BPF does not allow loops, so it is not Turing complete. A few years ago, an extension to the original BPF was created: eBPF(Enhanced BPF) adding new data structures (hashtables, arrays), tail calls, kernel hooks, etc... Furthermore, nowadays BPF programs can be executed in multiple kernel hooks, like networking, kprobes or system calls.

If you want to deep more about how to load BPF programs in your kernel I recommend Cilium docs, Brendan Gregg talks and bcc project.


XDP means Xpress Data Path; this technology was introduced in the kernel to make the packet processing much faster, so with XDP you can drop/pass packets very close to the network driver, without getting the packet into the kernel userspace. This technology made Facebook Load Balancers 10x faster than IPVS deployments.

To have a simple vision of this technology, the following image from Cilium is self-explanatory:

If you have doubts about Cilium, you should check Thomas Graf presentations, where you can learn the kernel technology on deep and how cilium implements it.

If you want to try Cilium, in the usage documentation you have some examples and a vagrant box created for testing purpose. However, if you want to test it using Kubernetes, things get a bit more complicated, you can use Minikube, but for testing networking tunnels you need more nodes and minikube just provides one.

I created a Terraform recipe where you can set-up a Kubernetes cluster with Cilium-cni on Google Cloud Platform in a few minutes. To start playing with that, you need to follow the process described in the Readme.

As soon that you login into the master node, you can see a few cilium daemon set pods working. If you want to know the cilium status you can run the following command:

for cilium in $(kubectl -n kube-system get pods --selector=k8s-app=cilium --output=jsonpath={.items..metadata.name}); do
        echo "===============${cilium}==============="
        kubectl -n kube-system exec $cilium cilium status

This command iterate over all cilium-agents pods, after that, it executes cilium status comand to know the status of the daemon set. The sample output is the following, where you can see the list of endpoints and the status of all 3-party integrations.


This command iterate over all cilium-agent pods
Allocated IPv4 addresses:
Allocated IPv6 addresses:
KVStore:            Ok   Etcd: - (Leader) 3.1.0
ContainerRuntime:   Ok
Kubernetes:         Ok   OK
Cilium:             Ok   OK

This setup is using VXLAN, a network virtualization technology that encapsulates Layer 2 frames within layer 4 UDP packets; Cilium also supports Geneve (If you want to deep more, you can read this post by Russel Bryant). To know the list of tunnels (The list of k8s nodes) that cilium is using, you can use the following command:

for cilium in $(kubectl -n kube-system get pods --selector=k8s-app=cilium --output=jsonpath={.items..metadata.name}); do
        echo "===============${cilium}==============="
        kubectl -n kube-system exec $cilium cilium bpf tunnel list

This command iterate over all cilium-agent pods, after that, it executes the command cilium bpf tunnel list where you can see the network assigned per each node, and the IP address of the node. The correct output should be similar to this:


When you try to schedule a new service, cilium will allocate a new IP to load balance the traffic, the list of the Load Balancers can be checked using the command cilium service list on any cilium-agent pod, the sample command is the following:

for cilium in $(kubectl -n kube-system get pods --selector=k8s-app=cilium --output=jsonpath={.items..metadata.name}); do
        echo "===============${cilium}==============="
        kubectl -n kube-system exec $cilium cilium service list

ID   Frontend            Backend
1       1 =>
2       1 =>
3   1 =>
                                                 2 =>
                                                 3 =>

From here, you can follow the policy enforcement guide from the cilium docs. In the coming months, I will share with you all the stuff I am learning.


Related articles: