Self-hosted Kubernetes

The self-hosted Kubernetes example provisions a 3 node "self-hosted" Kubernetes v1.6.4 cluster. bootkube is run once on a controller node to bootstrap Kubernetes control plane components as pods before exiting. An etcd3 cluster across controllers is used to back Kubernetes and coordinate Container Linux auto-updates (enabled for disk installs).

Requirements

Ensure that you've gone through the matchbox with rkt or matchbox with docker guide and understand the basics. In particular, you should be able to:

  • Use rkt or Docker to start matchbox
  • Create a network boot environment with coreos/dnsmasq
  • Create the example libvirt client VMs
  • /etc/hosts entries for node[1-3].example.com (or pass custom names to k8s-certgen)

Install bootkube v0.4.4 and add it somewhere on your PATH.

$ bootkube version
Version: v0.4.4

Examples

The examples statically assign IP addresses to libvirt client VMs created by scripts/libvirt. The examples can be used for physical machines if you update the MAC addresses. See network setup and deployment.

Assets

Download the CoreOS image assets referenced in the target profile.

$ ./scripts/get-coreos stable 1353.7.0 ./examples/assets

Add your SSH public key to each machine group definition as shown.

{
    "profile": "bootkube-worker",
    "metadata": {
        "ssh_authorized_keys": ["ssh-rsa pub-key-goes-here"]
    }
}

Use the bootkube tool to render Kubernetes manifests and credentials into an --asset-dir. Later, bootkube will schedule these manifests during bootstrapping and the credentials will be used to access your cluster.

$ bootkube render --asset-dir=assets --api-servers=https://node1.example.com:443 --api-server-alt-names=DNS=node1.example.com --etcd-servers=http://127.0.0.1:2379

Containers

Use rkt or docker to start matchbox and mount the desired example resources. Create a network boot environment and power-on your machines. Revisit matchbox with rkt or matchbox with Docker for help.

Client machines should boot and provision themselves. Local client VMs should network boot CoreOS and become available via SSH in about 1 minute. If you chose bootkube-install, notice that machines install CoreOS and then reboot (in libvirt, you must hit "power" again). Time to network boot and provision physical hardware depends on a number of factors (POST duration, boot device iteration, network speed, etc.).

bootkube

We're ready to use bootkube to create a temporary control plane and bootstrap a self-hosted Kubernetes cluster.

Secure copy the kubeconfig to /etc/kubernetes/kubeconfig on every node which will path activate the kubelet.service.

for node in 'node1' 'node2' 'node3'; do
    scp assets/auth/kubeconfig core@$node.example.com:/home/core/kubeconfig
    ssh core@$node.example.com 'sudo mv kubeconfig /etc/kubernetes/kubeconfig'
done

Secure copy the bootkube generated assets to any controller node and run bootkube-start.

$ scp -r assets core@node1.example.com:/home/core
$ ssh core@node1.example.com 'sudo mv assets /opt/bootkube/assets && sudo systemctl start bootkube'

Optionally watch the Kubernetes control plane bootstrapping with the bootkube temporary api-server. You will see quite a bit of output.

$ ssh core@node1.example.com 'journalctl -f -u bootkube'
[  299.241291] bootkube[5]:     Pod Status:     kube-api-checkpoint     Running
[  299.241618] bootkube[5]:     Pod Status:          kube-apiserver     Running
[  299.241804] bootkube[5]:     Pod Status:          kube-scheduler     Running
[  299.241993] bootkube[5]:     Pod Status: kube-controller-manager     Running
[  299.311743] bootkube[5]: All self-hosted control plane components successfully started

You may cleanup the bootkube assets on the node, but you should keep the copy on your laptop. It contains a kubeconfig used to access the cluster.

Verify

Install kubectl on your laptop. Use the generated kubeconfig to access the Kubernetes cluster. Verify that the cluster is accessible and that the kubelet, apiserver, scheduler, and controller-manager are running as pods.

$ KUBECONFIG=assets/auth/kubeconfig
$ kubectl get nodes
NAME                STATUS    AGE
node1.example.com   Ready     3m
node2.example.com   Ready     3m
node3.example.com   Ready     3m

$ kubectl get pods --all-namespaces
NAMESPACE     NAME                                       READY     STATUS    RESTARTS   AGE
kube-system   checkpoint-installer-p8g8r                 1/1       Running   1          13m
kube-system   kube-apiserver-s5gnx                       1/1       Running   1          41s
kube-system   kube-controller-manager-3438979800-jrlnd   1/1       Running   1          13m
kube-system   kube-controller-manager-3438979800-tkjx7   1/1       Running   1          13m
kube-system   kube-dns-4101612645-xt55f                  4/4       Running   4          13m
kube-system   kube-flannel-pl5c2                         2/2       Running   0          13m
kube-system   kube-flannel-r9t5r                         2/2       Running   3          13m
kube-system   kube-flannel-vfb0s                         2/2       Running   4          13m
kube-system   kube-proxy-cvhmj                           1/1       Running   0          13m
kube-system   kube-proxy-hf9mh                           1/1       Running   1          13m
kube-system   kube-proxy-kpl73                           1/1       Running   1          13m
kube-system   kube-scheduler-694795526-1l23b             1/1       Running   1          13m
kube-system   kube-scheduler-694795526-fks0b             1/1       Running   1          13m
kube-system   pod-checkpointer-node1.example.com         1/1       Running   2          10m

Try deleting pods to see that the cluster is resilient to failures and machine restarts (CoreOS auto-updates).

Going further

Learn to upgrade a self-hosted Kubernetes cluster.