Free to read · CKA · Cluster architecture and kubectl fluency

Demo: a guided tour of a kubeadm cluster

How the control plane and the nodes of a kubeadm cluster work together, how the API is organised, and the kubectl and exam habits that make every later module faster.

Demo · 15 min. This module is free on this site. The rest of the course, the quizzes and the labs open with a free account.

What you will do

You will walk around a two-node kubeadm cluster (controlplane and node01) and find every component from the previous lessons with your own eyes: the static pods, the kubelet, the runtime, the CNI and etcd. Run the commands on a lab cluster as you read. Your names, IPs, versions and IDs will differ; the shape of the output will not.

Start a lab such as C12 to get a cluster, then work on controlplane as root.

Step 1: the nodes

kubectl get nodes -o wide
NAME           STATUS   ROLES           AGE   VERSION   INTERNAL-IP   OS-IMAGE             CONTAINER-RUNTIME
controlplane   Ready    control-plane   42m   v1.37.1   10.0.0.10     Ubuntu 24.04.3 LTS   containerd://2.1.4
node01         Ready    <none>          39m   v1.37.1   10.0.0.11     Ubuntu 24.04.3 LTS   containerd://2.1.4

The CONTAINER-RUNTIME column already tells you the CRI implementation. The control plane node carries a taint that keeps normal pods away:

kubectl describe node controlplane | grep -A1 Taints
Taints:             node-role.kubernetes.io/control-plane:NoSchedule
Unschedulable:      false

Step 2: the system pods

kubectl get pods -n kube-system -o wide
NAME                                       READY   STATUS    RESTARTS   AGE   IP             NODE
calico-kube-controllers-6c8d5f9b7-qk2lp    1/1     Running   0          41m   192.168.49.1   controlplane
calico-node-7xgkd                          1/1     Running   0          39m   10.0.0.11      node01
calico-node-fz9wm                          1/1     Running   0          41m   10.0.0.10      controlplane
coredns-5d78c9869d-8vdrn                   1/1     Running   0          42m   192.168.49.2   controlplane
coredns-5d78c9869d-wl4sp                   1/1     Running   0          42m   192.168.49.3   controlplane
etcd-controlplane                          1/1     Running   0          42m   10.0.0.10      controlplane
kube-apiserver-controlplane                1/1     Running   0          42m   10.0.0.10      controlplane
kube-controller-manager-controlplane       1/1     Running   0          42m   10.0.0.10      controlplane
kube-proxy-hn6tq                           1/1     Running   0          39m   10.0.0.11      node01
kube-proxy-p2w8c                           1/1     Running   0          42m   10.0.0.10      controlplane
kube-scheduler-controlplane                1/1     Running   0          42m   10.0.0.10      controlplane

Read the names. Pods ending in -controlplane are mirror pods of static pods. Pods with random suffixes belong to Deployments (coredns, calico-kube-controllers) or DaemonSets (calico-node, kube-proxy, one per node). Pods that use the node's IP run with hostNetwork: true.

Confirm who owns what:

kubectl -n kube-system get pod kube-apiserver-controlplane -o jsonpath='{.metadata.ownerReferences[0].kind}{"\n"}'
kubectl -n kube-system get pod kube-proxy-hn6tq -o jsonpath='{.metadata.ownerReferences[0].kind}{"\n"}'
Node
DaemonSet

A mirror pod is owned by the Node. That is the giveaway for a static pod.

Step 3: the static pod manifests

ls /etc/kubernetes/manifests
grep staticPodPath /var/lib/kubelet/config.yaml
etcd.yaml  kube-apiserver.yaml  kube-controller-manager.yaml  kube-scheduler.yaml
staticPodPath: /etc/kubernetes/manifests

Pull out the most useful API server flags:

grep -E -- '--(etcd-servers|service-cluster-ip-range|authorization-mode|enable-admission-plugins)' \
  /etc/kubernetes/manifests/kube-apiserver.yaml
    - --authorization-mode=Node,RBAC
    - --enable-admission-plugins=NodeRestriction
    - --etcd-servers=https://127.0.0.1:2379
    - --service-cluster-ip-range=10.96.0.0/12

--etcd-servers pointing at 127.0.0.1 tells you etcd runs on this same node: a stacked topology. An external etcd would show other hosts here, and there would be no etcd.yaml.

Step 4: the kubelet

systemctl is-active kubelet; systemctl is-enabled kubelet
ps -o args= -C kubelet | tr ' ' '\n' | grep -- '--'
active
enabled
--bootstrap-kubeconfig=/etc/kubernetes/bootstrap-kubelet.conf
--kubeconfig=/etc/kubernetes/kubelet.conf
--config=/var/lib/kubelet/config.yaml

Your flag list may be shorter or longer; kubeadm moves settings between flags and the config file over time. The --config file holds the rest:

grep -E 'cgroupDriver|containerRuntimeEndpoint|clusterDNS' -A1 /var/lib/kubelet/config.yaml
cgroupDriver: systemd
clusterDNS:
- 10.96.0.10
containerRuntimeEndpoint: unix:///run/containerd/containerd.sock

Step 5: the runtime

crictl version
crictl ps --name kube-apiserver
Version:  0.1.0
RuntimeName:  containerd
RuntimeVersion:  v2.1.4
RuntimeApiVersion:  v1

CONTAINER       IMAGE           CREATED          STATE     NAME             ATTEMPT   POD ID          POD
3f1b9e0c7d2a1   a4b2c8d9e1f00   42 minutes ago   Running   kube-apiserver   0         9c0d2e7f1a3b4   kube-apiserver-controlplane

This is the same API server you saw in kubectl, seen from below. If the API server ever stops answering, this is where you look.

Step 6: the pod network

ls /etc/cni/net.d
kubectl get nodes -o custom-columns=NAME:.metadata.name,PODCIDR:.spec.podCIDR
10-calico.conflist  calico-kubeconfig
NAME           PODCIDR
controlplane   192.168.0.0/24
node01         192.168.1.0/24

The first config file in sort order is the one the runtime loads. Calico assigns pod IPs from its own IP pools inside the cluster's pod CIDR, so pod IPs you see may not fall inside the node's podCIDR slice. That is normal for Calico; other plugins use the slice directly.

Step 7: etcd membership

export ETCDCTL_API=3
alias e='etcdctl --endpoints=https://127.0.0.1:2379 --cacert=/etc/kubernetes/pki/etcd/ca.crt \
  --cert=/etc/kubernetes/pki/etcd/server.crt --key=/etc/kubernetes/pki/etcd/server.key'
e member list -w table
e endpoint status -w table
+------------------+---------+--------------+------------------------+------------------------+------------+
|        ID        | STATUS  |     NAME     |       PEER ADDRS       |      CLIENT ADDRS      | IS LEARNER |
+------------------+---------+--------------+------------------------+------------------------+------------+
| 8e9e05c52164694d | started | controlplane | https://10.0.0.10:2380 | https://10.0.0.10:2379 |      false |
+------------------+---------+--------------+------------------------+------------------------+------------+
+------------------------+------------------+---------+---------+-----------+-----------+------------+
|        ENDPOINT        |        ID        | VERSION | DB SIZE | IS LEADER | RAFT TERM | RAFT INDEX |
+------------------------+------------------+---------+---------+-----------+-----------+------------+
| https://127.0.0.1:2379 | 8e9e05c52164694d |  3.7.3  |  5.2 MB |      true |         2 |       8131 |
+------------------------+------------------+---------+---------+-----------+-----------+------------+

(Output trimmed to the most useful columns; your etcd version follows your Kubernetes version.) One member, and it is the leader. In an HA cluster you would see three or five rows, each with its own peer URL, and exactly one leader.

Step 8: put it together

You have now located every piece:

PieceWhere you found it
Control planeStatic pods from /etc/kubernetes/manifests, owned by the Node
etcd topology--etcd-servers and etcd.yaml: stacked
kubeletsystemd, /var/lib/kubelet/config.yaml
CRIcontainerd via unix:///run/containerd/containerd.sock
CNICalico, /etc/cni/net.d/10-calico.conflist
Service CIDR--service-cluster-ip-range on the API server

Common mistakes

  • Running etcdctl without the certificates. It hangs, then fails with a context deadline error.
  • Reading /etc/cni/net.d on one node and assuming the other node matches. Check the node the task names.
  • Trusting the example values in this demo. Always read your own cluster.

Practise it

Write your findings up as reports, exactly as an exam task would ask:

Practise this on a real cluster in your account: C12, C36:a, C28:a.

Official documentation