Welcome to the Drills
You run Kubernetes clusters and want to understand what happens between the Pods: who hands out the addresses, how a Service IP turns into a Pod, why DNS hangs, how traffic comes in from outside and goes back out, how you block it, encrypt it and observe it. That is exactly what the CKNE tests. Not as multiple choice, but in the terminal, on real clusters, against the clock.
The curriculum is vendor-neutral, but the book still has a common thread: Cilium. It replaces kube-proxy with eBPF, enforces NetworkPolicies up to Layer 7 and brings a Gateway API controller, BGP, Egress Gateway, ClusterMesh and, with Hubble, visibility into every flow. Where the curriculum asks for standards (NetworkPolicy, Gateway API, CoreDNS, EndpointSlices, MCS API), you practice the standard – and see how Cilium implements it.
This book assumes CKA knowledge – just like the real exam. You should
be able to create Pods, Deployments and Services, read YAML and find
faults on nodes with kubectl, crictl,
journalctl and ssh. Everything
network-specific you learn here.
How the book works
- One day, one topic. Each day has a goal, a short briefing with the curriculum domain and 5–8 drills. You type every command yourself and compare with “Expected”.
- Before, configure, after. Every feature starts with
the state without it and ends with proof:
curl,tcpdump,hubble observeorcilium-dbgin the agent on the right node. - Imperative first. You generate YAML with
k create … $do > datei.yamlinstead of typing it by hand. Before every change to Cilium you back up the Helm values and restore them at the end of the day – unless a later day needs them; the Checkpoint then says so. - Speed Drills at the end of each day: small exam tasks with a target time.
- Checkpoint, Challenge, Exam Tip, Enterprise Note close out every day.
- One application through many days: notes-api, a small Go API. You expose it as a NodePort, lock it in with L7 policies and require authentication per identity. Alongside it, agnhost, echo-basic and netshoot serve as tools.
- Competition (Day 21): a broken network that you repair under time pressure – all curriculum domains mixed.
- Appendix A is the cheat sheet, Appendix B a practice exam: two hours, about 17 tasks, weighted by the curriculum.
What you need
Five VMs with Ubuntu 24.04 on the same network, with SSH access and
sudo. You need the first four from Day 0, west
only from Day 17:
| Role | Purpose | IP in the book | Minimum |
|---|---|---|---|
| cp1 | control plane, working host | 172.16.0.126 | 4 vCPU, 11 GB RAM |
| worker1, worker2 | workers of cluster ost |
172.16.0.228, .83 | 2 vCPU, 3 GB each |
| router | client/server outside, BGP | 172.16.0.123 | 2 vCPU, 3 GB |
| west | second cluster, from Day 17 | 172.16.0.182 | 4 vCPU, 4 GB |
cp1 is that large because Prometheus and Grafana also
run there from Day 19. In the lab, the VM for router is
called prod and the one for west is called
argo – name yours however you like, and use your own names
and IPs everywhere. There is no name resolution between the VMs; that is
why the book uses IPs everywhere.
Proxmox, VMware, a cloud or Multipass – it does not matter. On your own machine, Multipass is the easiest way. So that Day 0 finds your SSH key on all VMs, you pass it in via cloud-init:
cat > ckne-cloud-init.yaml <<EOF
#cloud-config
ssh_authorized_keys:
- $(cat ~/.ssh/id_ed25519.pub)
EOF
multipass launch 24.04 --name cp1 --cpus 4 --memory 11G --disk 40G \
--cloud-init ckne-cloud-init.yaml
multipass launch 24.04 --name worker1 --cpus 2 --memory 3G --disk 20G \
--cloud-init ckne-cloud-init.yaml
multipass launch 24.04 --name worker2 --cpus 2 --memory 3G --disk 20G \
--cloud-init ckne-cloud-init.yaml
multipass launch 24.04 --name router --cpus 2 --memory 3G --disk 10G \
--cloud-init ckne-cloud-init.yaml
multipass list
# not until Day 17:
# multipass launch 24.04 --name west --cpus 4 --memory 4G --disk 20G \
# --cloud-init ckne-cloud-init.yamlmultipass list shows the IPs; you create
west only on Day 17. Together, the four VMs need about 20
GB of RAM – plan your machine accordingly. In the lab, the network
interface is called eth0; if yours has a different name
(ip -br link), use its name everywhere the book writes
eth0.
On a Mac with Apple Silicon the VM is arm64. That is why all download
commands detect the architecture themselves
(ARCH=$(dpkg --print-architecture)), and all images in the
book exist for amd64 and arm64. File names and output then show
arm64 where the book shows amd64.
Time: Days 0 to 20 add up to just under 28 hours (1665 minutes), plus the Competition and the two-hour practice exam. At one training day per evening and five evenings per week, you are through in a little over four weeks.
On Day 0 you use kubeadm to build the cluster ost
running Kubernetes v1.37.2 – deliberately
without kube-proxy. Cilium takes over the Pod network
(IPAM cluster-pool 10.0.0.0/16, VXLAN tunnel), Services
(kube-proxy replacement) and Hubble. On top come only metrics-server and
local-path as the StorageClass. Everything else – Multus, Gateway API
CRDs, cert-manager, FRR, SPIRE, Prometheus – is installed by the day
that teaches it. Whatever a day changes cluster-wide, it rolls back at
the end, unless a later day needs it; the Checkpoint then says so.
The companion repo
The source code of notes-api (folder notes-api) and the
material for the book live at
https://github.com/thomaszachmann/cilium-zum-anfassen. You
do not need to build anything: the image
ghcr.io/thomaszachmann/notes-api:0.5.1 is ready-made for
amd64 and arm64. You still type the manifests and policies in the book
yourself – that is the training.
The exam in brief
| Domain (CKNE curriculum) | Weight | Days |
|---|---|---|
| Core Infrastructure and CNI | 15 % | 2, 3, 4, 5 |
| Service Networking and DNS | 25 % | 6, 7, 8, 9, 17 |
| Advanced Traffic Management | 20 % | 10, 11, 15, 16, 18 |
| Network Security and Policy | 25 % | 12, 13, 14 |
| Observability | 15 % | 19, 20 |
Days 0 and 1 prepare the training ground and your craft, Day 21 mixes all domains.
As things stand, the exam takes two hours, runs online and proctored
and consists of practical tasks in a remote desktop environment with a
terminal and a browser. CKA knowledge is assumed, even though the CKA is
formally not a prerequisite. Format, passing score and the permitted
documentation pages change occasionally – the authoritative source is
the information from the Linux Foundation at
https://training.linuxfoundation.org. You can find the
curriculum at https://github.com/cncf/curriculum.
Note: This book is an independent practice book. It is not affiliated with the Linux Foundation or the Cloud Native Computing Foundation and is neither reviewed nor endorsed by them. “Certified Kubernetes Network Engineer” and “CKNE” are trademarks of the Linux Foundation.
Versions
| Component | Version | from Day |
|---|---|---|
| Ubuntu / kernel | 24.04.5 LTS / 6.8.0-149-generic | 0 |
| Kubernetes (kubeadm, kubelet, kubectl) | v1.37.2 | 0 |
| containerd / runc / cri-tools | 2.4.2 / 1.5.3 / 1.37.0 | 0 |
| Cilium (Helm chart) / cilium-envoy | 1.20.5 / v1.39.6 | 0 |
| cilium CLI / hubble CLI | v0.20.3 / v1.20.4 | 0 |
| Helm | v4.3.2 | 0 |
| metrics-server / local-path-provisioner | v0.9.1 / v0.0.37 | 0 |
| Multus CNI (thick) | v4.3.2 | 5 |
| CoreDNS | v1.14.6 | 8 |
| Gateway API CRDs (standard channel) | v1.6.3 | 10 |
| cert-manager / cmctl | v1.21.3 / v2.6.2 | 11 |
| SPIRE (from the Cilium chart) | 1.15.5 | 14 |
FRR (Ubuntu package on router) |
8.4.4 | 15 |
| Gateway API Inference Extension / llm-d Router | v1.6.2 / v1.0.0 | 18 |
| agentgateway / llm-d-inference-sim | 1.7.0 / v1.2.0 | 18 |
| kube-prometheus-stack (chart) | 93.4.1 | 19 |
| Jaeger / HotROD | 2.23.0 / 2.23.0 | 20 |
Test images: notes-api 0.5.1, agnhost 2.66.1, nicolaka/netshoot
v0.16, nginx 1.29, curlimages/curl 8.17.0, echo-basic v1.5.1, grpcurl
v1.9.3. All output comes from real runs, shortened where it makes sense.
Values such as the API key lab-key-123 or the Grafana
password ckne-lab are lab values only.
Enjoy your drills.