TKS (TJ's Kubernetes Service)

Summary

Declaratively build a 4 node Kubernetes cluster on Proxmox using Ansible and QEMU. Optionally enable advanced features including ingress, load balancing, unattended upgrades etc.

Approximate deployment time: 25 minutes

Requirements

1. Proxmox server
2. DNS Server*
3. Ansible 2.7.0+. Known incompatibility with a previous build.
   _{*A DNS server is not technically required, it is possible to manually add entries corresponding to your node hostnames to your Proxmox's hosts file.}

Instructions

Required:

1. Modify the vars.yml file with values specific to your environment.
2. Provision DNS A records for the IP Addresses & Hostnames you defined for your nodes in the vars.yml file.
3. Modify the inventory.ini file to reflect your chosen DNS records and the location of the SSH keys used to connect to the nodes.
4. Run the deployment: ansible-playbook -i inventory.ini site.yml
5. After deployment, a ~/.kube directory will be created on your workstation. Within your config an authentication_token file can be be found. This token is used to authenticate against the Kubernetes API and Dashboard using your account. To connect to the dashboard, install kubectl on your workstation and run kubectl proxy then navigate to the Dashboard Endpoint in your browser.

Optional:

To enable an optional feature, fill in the additional parameters in vars.yml and execute a playbook listed below.

Feature	Command	Requirements
NFS backed persistent storage	ansible-playbook -i inventory.ini playbooks/optional/deploy_nfs_provisioner.yml
MetalLB Load Balancer	ansible-playbook -i inventory.ini playbooks/optional/deploy_metallb.yml
NGINX Ingress Controller	ansible-playbook -i inventory.ini playbooks/optional/deploy_ingress-nginx.yml	MetalLB or other Load Balancer integration
DataDog agents	ansible-playbook -i inventory.ini playbooks/optional/deploy_datadog.yml
Unattended Upgrades	ansible-playbook -i inventory.ini playbooks/optional/enable_unattended_upgrades.yml

Tips

1. You can rollback the entire deployment with: ansible-playbook -i inventory.ini playbooks/optional/delete_all_resources.yml
2. If Calico isn't deploying correctly it's likely the CIDR you assigned to it in vars.yml conflicts with your network.
3. There appears to be an issue with Proxmox's cloud-init implementation. Perhaps just with Debian? As a result, your VM might not have the correct information in /etc/resolv.conf and may also have multiple IP Addresses assigned to the eth0 network interfaces. Furthermore, if you do not have a DHCP server active in the network that you are provisoning the VMs to, it is entirely possible that nothing will be present at all in /etc/resolv.conf.
4. See this repository to do this with LXC instead. Benefits of using LXC include:

* No virtualization overhead means better performance
* Ability to directly mount volumes from your server into your containers.

TODO

1. Add better support for multi-node Proxmox clusters.
2. Perform security audit and enhance if necessary.
3. Add info to README about updating inventory file and how to handle SSH key generation and propegation.
4. Create playbook to upgrade kubernetes version for kubeadm cluster.
5. Move dashboard deployment to optional features.

Problems

1. The proxmox_kvm module is out of date and does not support cloudinit related api calls. Meaning shell commands must be used instead to perform qm create tasks.
2. The k8s module does not support applying Kubernetes Deployments from URL. Instead of using get_url to download them first, and then apply them with k8s, I just use shell to run a kubectl apply -f. Feature Request here.
3. Miscellaneous qcow2 image issues:

OS	Issue
CentOS	A nameserver is baked into /etc/resolv.conf by default. Bug Report here
CoreOS	Proxmix issued cloud-init does not seem to configure networking properly.