README

This folder contains a Vagrantfile to set up a local kubernetes cluster. Ansible is used for provisioning (see playbooks).

Quick start

The cluster nodes are created by running vagrant up --provider libvirt.

You can then run the playbooks to install Kubernetes from the host, do it like this:

ansible-playbook playbooks/provision-all.yml

There is no need to specify the inventory since it is already configured in ansible.cfg.

Once ansible is finished provisioning you can access the cluster from the control-plane1 node or from the nfs server. Example:

$ vagrant ssh nfs
$ kubectl get nodes
NAME             STATUS   ROLES           AGE     VERSION
control-plane1   Ready    control-plane   4m42s   v1.24.2
control-plane2   Ready    control-plane   3m21s   v1.24.2
control-plane3   Ready    control-plane   98s     v1.24.2
worker1          Ready    <none>          64s     v1.24.2
worker2          Ready    <none>          77s     v1.24.2

Requirements

If you want to use vagrant with the libvirt provider, make sure you set up libvirt with KVM first and install the vagrant-libvirt plugin. Instructions for setting up libvirt with KVM on Ubuntu can be found here. The vagrant-libvirt plugin can be found here.

Architecture

Vagrant is used to create virtual machines according to the specification in the Vagrantfile. These machines are then provisioned using ansible playbooks to form a Kubernetes cluster, NFS server and load balancer. The architecture is similar to this diagram.

Note that the default configuration only has a single control-plane node.

Options

There are multiple ways to configure the vagrant cluster. This section describes some of the possibilities, but by no means all of them.

Check the examples folder for some example configurations. It contains inventory files ending with .ini and vagrant files starting with Vagrantfile. If you wish to try one of the examples, replace the Vagrantfile with the example vagrant file (e.g. mv examples/Vagrantfile.ha Vagrantfile). It is already configured to use the corresponding example inventory.

Operating system

You can choose between Ubuntu and CentOS as operating system for the nodes. Simply uncomment the vagrant box you want to use in the Vagrantfile:

config.vm.box = "generic/centos9s"
# config.vm.box = "generic/ubuntu2204"

OR add the box in the hosts hash (see worker2 and worker3):

hosts = {
    "control-plane1" => { "memory" => 1536, "ip" => "192.168.10.10"},
    "worker1" => { "memory" => 1536, "ip" => "192.168.10.30"},
    "worker2" => { "memory" => 1536, "ip" => "192.168.10.31", "box" => "generic/ubuntu2204"},
    "worker3" => { "memory" => 1024, "ip" => "192.168.10.32", "box" => "generic/ubuntu2204"},
    "nfs" => { "memory" => 512, "ip" => "192.168.10.20"}
}

High availability control plane

The provided playbooks support setting up a HA control plane for kubernetes. You will need at least three control-plane nodes configured both in the Vagrantfile and in the inventory file. Additionally, you probably want to set up an external client instead of relying on one of the control-plane machines for talking to the API. You can for example use the nfs/loadbalancer machine. See examples/ha.ini and examples/Vagrantfile.ha for an example on how to configure this.

Ops Scenarios

A number of scenarios with instructions and commands are available in the scenarios folder.

Test application: wordpress

The vagrant environment contains a wordpress deployment to simulate a workload. Add 192.168.10.40 example.com wordpress.example.com to /etc/hosts in order to access wordpress at wordpress.example.com. Or maybe set up dnsmasq to resolve the whole kube-ops domain to the IP address of the ingress controller.

See playbooks/templates/wordpress-values.yaml.j2 for configuration values. The login credentials are kube-ops:p455w0rd at http://wordpress.example.com/wp-admin.

There are a few other components deployed in the cluster as base infrastructure.

nfs-client-provisioner for persistent volume provisioning
nginx-ingress as ingress controller
metallb as load balancer

Here is a diagram describing how a request flows through the system.

The metalLB is configured to use IP addresses from 192.168.10.40 to 192.168.10.50. Since the nginx ingress controller is the first and only application in the cluster with a service of type LoadBalancer, it will get the IP address 192.168.10.40.

The NFS provisioner is responsible for creating persistent volumes from the NFS server. It polls the API server for new persistent volume claims and creates a new volume as soon as it detects a new claim.

Issues to be aware of

There is a bug that causes pods to remain in unknown state on nodes that are not ready, if they have volumes attached.
- shutdown taint issue
Helm does not automatically update repository information. Make sure you get the version you want of all charts, and if needed run helm repo update.
Be careful what you wish for: pod disruption budgets won't stop nodes from going down, but they will prevent you from evicting pods!
Backing up only etcd data and root certificate is not enough. Old secrets (tokens) will not be valid and cause applications to fail. (Kube-proxy is one example.)
Coredns is running on the control-plane at first. This will make all pods loose DNS if/when the control-plane goes down. If the coredns pods are moved to some other node (by killing them after other nodes have joined) all applications will continue to operate normally when the control-plane goes down.
Vagrant may occasionally try to run the ansible provisioner for all VMs. If this happens, just Ctrl + C and start it again with vagrant provision. (You'll notice this by looking for duplicate tasks and output.)

lentzi90 / kube-ops