xiaobaowan1988 / ansible_kvm_k8s

Deploying a HA-Ready Kubernetes Cluster on KVM with Ansible

This repo provides code for an automated way of deploying a K8S cluster with Ansible on KVM hypervisor. The project can be divided in three phases. First, cloud image is downloaded and customized accordingly using libguestfs popular tool virt-customize. Second, using the libvirt module and XML definition files, VMs were started on KVM hypervisor. And third, initialization scripts on each VM were run to deploy the K8S cluster. A medium story with more detailed discussion is here. --> https://medium.com/@emrah-t/kubernetes-cluster-up-and-running-in-seconds-with-ansible-devops-diaries-757374284f32

Prerequisites

1. A Linux computer (VM or physical) with the required KVM/libvirt components installed (kvmhost). --> https://linuxhint.com/install-kvm-ubuntu-22-04/
2. Ansible binaries installed on the controller machine. --> https://docs.ansible.com/ansible/latest/installation_guide/installation_distros.html
3. An intermediate level of domain (Ansible, KVM/libvirt, Kubernetes) knowledge.

Code

1. Ansible roles were employed to keep tasks simpler and more organized.
2. The Ubuntu 20.04 cloud image is downloaded and customized accordingly using virt-customize.
3. As for customization, configuration files (under kvmhost_scripts directory) and SSH keys were copied, root passwords were set.
4. community.libvirt.virt module was used to manage KVM VMs.
5. XML definition files under templates directory were used to boot up the VMs.
6. Four VMs were defined (HAProxy Load Balancer, master and two workers).
7. After starting the VMs, initialization scripts (under kvmhost_scripts directory) were run to bootstrap K8S cluster.
8. Resulting topology is given below.

Execution

1. Clone the repo.
2. Go to directory playbook.yml stands and run "ansible-playbook playbook.yml -i inventory/hypervisors.yml"
3. Follow the creation of VMs via virsh or Virtual Machine Manager. After seeing they're up, connect with console or SSH and follow logs in /tmp/init.log (screencast below).
4. Final check with "kubectl get nodes" on the master node (screencast below). All bootstrapping lasts around 350 seconds on my home hardware.

screencast.webm