Skip to content
Really cheap Kubernetes cluster on AWS with kubeadm
Branch: master
Clone or download
Pull request Compare This branch is 58 commits ahead of upmc-enterprises:master.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
examples Add cert-manager for TLS certificates Dec 16, 2018
manifests Add cert-manager for TLS certificates Dec 16, 2018
.gitignore EBS persistent volumes and stop accidental k8s upgrades Aug 26, 2018
LICENSE Initial commit Sep 28, 2016
README.md Simplify example command Mar 6, 2019
main.tf Generate the bootstrap token with terraform to fix #5 Mar 6, 2019
master.sh Add cert-manager for TLS certificates Dec 16, 2018
outputs.tf
variables.tf
worker.sh

README.md

Really cheap Kubernetes cluster on AWS with kubeadm

This repository contains a bunch of Bash and Terraform code which provisions what I believe to be the cheapest possible single master Kubernetes cluster on AWS. You can run a 1 master, 1 worker cluster for somewhere around $6 a month, or just the master node (which can also run pods) for around $3 a month.

To achieve this, it uses m1.small spot instances and the free ephemeral storage they come with instead of EBS volumes.

Current features:

  • Automatic backup and recovery. So if your master gets terminated, when the replacement is provisioned by AWS it will pick up where the old one left off without you doing anything. 😁
  • Completely automated provisioning through Terraform and Bash.
  • Variables for many things including number of workers (provisioned using an auto-scaling group) and EC2 instance type.
  • Helm Tiller (currently v2.12.0)
  • External DNS and Nginx Ingress as a cheap ELB alternative, with Cert Manager for TLS certificates via Let's Encrypt.
  • Auto Scaling of worker nodes, if you enable the Cluster AutoScaler.
  • Persistent Volumes using GP2 storage on EBS.

Please use the releases rather than pulling from master. Master may be untested at any given point in time. This isn't designed for production (unless you're very brave) but I've found no stability issues so far for my personal development usage. However I have had instances where there is no available spot capacity for my chosen instance type in my Availability Zone which means you are without any nodes for a while...

Run it!

  1. Clone the repo
  2. Install Terraform
  3. Make an SSH key on us-east-1 from the AWS console
  4. Run terraform plan: terraform plan -var k8s-ssh-key=<aws-ssh-key-name> -var admin-cidr-blocks="<my-public-ip-address>/32"
  5. Build out infrastructure: terraform apply -var k8s-ssh-key=<aws-ssh-key-name> -var admin-cidr-blocks="<my-public-ip-address>/32"
  6. SSH to K8S master and run something: ssh ubuntu@$(terraform output master_dns) -i <aws-ssh-key-name>.pem kubectl get no
  7. If you enabled cert-manager, the Cert Manager Issuer for Let's Encrypt has been applied to the default namespace. You will also need to apply it to any other namespaces you want to obtain TLS certificates for.
  8. Done!

Optional Variables:

  • min-worker-count - The minimum size of the worker node Auto-Scaling Group (1 by default)
  • max-worker-count - The maximum size of the worker node Auto-Scaling Group (1 by default)
  • region - Which AWS region to use (us-east-1 by default)
  • az - Which AWS availability zone to use (a by default)
  • kubernetes-version - Which Kubernetes/kubeadm version to install (1.13.4 by default)
  • master-instance-type - Which EC2 instance type to use for the master node (m1.small by default)
  • master-spot-price - The maximum spot bid for the master node ($0.01 by default)
  • worker-instance-type - Which EC2 instance type to use for the worker nodes (m1.small by default)
  • worker-spot-price - The maximum spot bid for worker nodes ($0.01 by default)
  • cluster-name - Used for naming the created AWS resources (k8s by default)
  • backup-enabled - Set to "0" to disable the automatic etcd backups (1 by default)
  • backup-cron-expression - A cron expression to use for the automatic etcd backups (*/15 * * * * by default)
  • external-dns-enabled - Set to "0" to disable ExternalDNS (1 by default) - Existing Route 53 Domain required
  • nginx-ingress-enabled - Set to "1" to enable Nginx Ingress (0 by default)
  • nginx-ingress-domain - The DNS name to map to Nginx Ingress using External DNS ("" by default)
  • cert-manager-enabled - Set to "1" to enable Cert Manager (0 by default)
  • cert-manager-email - The email address to use for Let's Encrypt certificate requests ("" by default)
  • cluster-autoscaler-enabled - Set to "1" to enable the cluster autoscaler (0 by default)
  • k8stoken - Override the automatically generated cluster bootstrap token

Examples

Ingress Notes

As hinted above, this uses Nginx Ingress as an alternative to a Load Balancer. This is done by exposing ports 443 and 80 directly on each of the nodes (Workers and the Master) using a NodePort type Service. Unfortunately External DNS doesn't seem to work with Nginx Ingress when you expose it in this way, so I've had to just map a single DNS name (using the nginx-ingress-domain variable) to the NodePort service itself. External DNS will keep that entry up to date with the IPs of the nodes in the cluster; you will then have to manually add CNAME entries for your individual services.

I am well aware that this isn't the most secure way of exposing services, but it's secure enough for my purposes. If anyone has any suggestions on a better way of doing this without shelling out $20 a month for an ELB, please open an Issue!

Contributing

I've written this as a personal project and will do my best to maintain it to a good standard, despite having very limited free time. I very much welcome contributions in the form of Pull Requests and Issues (for both bugs and feature requests).

Note about the license

I am not associated with UPMC Enterprises, but because this project started off as a fork of their code I am required to leave their license in place. However this is still Open Source and so you are free to do more-or-less whatever you want with the contents of this repository.

You can’t perform that action at this time.