cover image

February 20, 2022

A Vercel-like PaaS beyond Jamstack with Kubernetes and GitOps, part I

Cluster setup

This article is the first part of the A Vercel-like PaaS beyond Jamstack with Kubernetes and GitOps series.

A Vercel-like PaaS beyond Jamstack with Kubernetes and GitOps

As told in the introduction, this part is about building a cheap, easy to rebuild Kubernetes cluster to get the ball rolling. I'd like to test things on a bare setup before using managed clusters such as AKS, EKS, GKE, etc.

  1. Start a fresh server
  2. Install k0s
  3. Install Lens and add k0s cluster
  4. Add cert-manager and nginx-ingress-controller
  5. An overview of how network traffic flows
  6. Next step

To run this setup I need a Linux system with at least 2 GB of RAM1, and a little more than the default 8 GB of disk space to make sure logs won't fill up all the available space. This is definitely not ideal but my goal here is to build a cheap setup.

I also need a wildcard subdomain pointing to this server. I'm using *

My server setup is:

  • 2GB RAM
  • 2 vCPUs
  • 16GB disk
  • Ubuntu 20.04

1. Start a fresh server

k0s can run on any server from any cloud provider as long as it can run a Linux distribution that is running either a Systemd or OpenRC init system.

AWS is my go-to provider, but going with any other cloud service provider shouldn't be a problem.

All aws commands I'm running on my workstation can be executed from AWS CloudShell or performed from the web Management Console.

First, I launch a t3.small EC2 instance running Ubuntu 20.04.

I'm providing my SSH key with the --key-name flag, and I'm attaching my EC2 instance to an existing security group that allows TCP traffic on ports 22, 80, 443 and 6443:

aws ec2 run-instances \
--image-id ami-04505e74c0741db8d \
--count 1 \
--instance-type t3.small \
--key-name k0s \
--block-device-mappings \
'DeviceName=/dev/sda1,Ebs={VolumeSize=16}' \
--security-group-ids sg-4327b00b \
--tag-specifications \

When I want to clean up everything and shutdown created instances, I use the --tag-specifications provided above to select instances and terminate them:

# store running instances ids in a variable
INSTANCES=`aws ec2 describe-instances \
--query \
Reservations[*].Instances[*].[InstanceId] \
--filters \
Name=tag:project,Values=k0sTest \
Name=instance-state-name,Values=running \
--output text`
# delete instances
aws ec2 terminate-instances --instance-ids $INSTANCES

Configure a DNS record to point a wildcard subdomain to the server

Since I'm using AWS Route53, I've made a script to speed up the operation.

I'm using it as follows to update the existing A record:

HOSTED_ZONE_ID=`aws route53 list-hosted-zones \
--query HostedZones[*].[Id,Name] \
--output text \
| grep | awk '{ print $1}'`
K0S_IP=`aws ec2 describe-instances \
--query \
Reservations[*].Instances[*].PublicIpAddress \
--filters \
Name=tag:project,Values=k0sTest \
Name=instance-state-name,Values=running \
--output text`
curl -sSlf \
| bash -s -- ${HOSTED_ZONE_ID} '\\' ${K0S_IP}

Now, any subdomain ending with, such as, will be routed to my EC2 instance.

This way I don't have to add a new CNAME record each time I create.

2. Install k0s

The $K0S_IP variable has already been set in section 1 and contains the server's IP address.

The private key I've attached to the server is in my Downloads folders and I use it to ssh to the server:

$ ssh -i ~/Downloads/k0s.pem ubuntu@$K0S_IP
Welcome to Ubuntu 20.04.3 LTS (GNU/Linux 5.11.0-1022-aws x86_64)

Once I'm connected to the EC2 instance, I can download the k0s binary and create a cluster configuration file:

# download k0s binary file:
curl -sSLf | sudo K0S_VERSION=v1.23.3+k0s.0 sh
# generate a default config file:
sudo k0s config create > k0s.yaml
# replace with server's public ip
# to grant access from the outside
PUBLIC_IP=`curl -s`
sed -i 's/^\( sans\:\)/\1\n - '$PUBLIC_IP'/g' k0s.yaml

The configuration file has been generated and I've replaced the EC2 instance's private IP address with its public IP address with the sed command to expose the Kubernetes API to the internet.

It's not a good practice, and in real life I'd prefer to use AWS VPN or my own OpenVPN setup to join the EC2 instance's network and query the Kubernetes API from the internal network.

Now, I can install a single node cluster:

$ sudo k0s install controller --single -c k0s.yaml
$ sudo k0s start
# wait a few seconds then:
$ sudo k0s status
Version: v1.23.3+k0s.0
Process ID: 3606
Role: controller
Workloads: true
SingleNode: true
# wait a minute then check if control-plane is up:
$ sudo k0s kubectl get nodes
No resources found
# not ready yet, wait and retry:
$ sudo k0s kubectl get nodes
ip-172-31-13-250 Ready control-plane 5s v1.23.3+k0s

The k0s cluster is now up and running. I check the server resources status to confirm it's not overloaded:

htop output command

3. Install Lens and connect to k0s cluster

Lens is a graphical UI for kubectl, it makes interacting with a cluster easier and debugging faster for me.

Once I've installed Lens, I skip the subscription process and add the cluster from File > Add ClusterFrom menu. It shows an input field where I can paste a user configuration.

lens config interface

Adding a new cluster to Lens

To get these credentials I go back to the server, then copy the whole yaml output of this command, and paste it in Lens:

$ sudo k0s kubeconfig admin \
| sed 's/'$(ip r | grep default | awk '{ print $9}')'/'$(curl -s'/g'
WARN[2022-02-05 18:49:43] no config file given, using defaults
apiVersion: v1
- cluster:
certificate-authority-data: LS0tLS1CRUdJTiBDRVJUS...
name: local
- context:
cluster: local
namespace: default
user: user
name: Default
current-context: Default
kind: Config
preferences: {}
- name: user
client-certificate-data: LS0tLS1CRUdJTiBDRVJUSUZJQ0...
client-key-data: LS0tLS1CRUdJTiBSU0EgUFJJVkFURSBLRV...

Now I'm able to connect to the cluster and show an overview of the workload in all namespaces:

lens cluster overview

Lens overview of the new cluster

4. Add nginx-ingress-nginx and cert-manager

To install third-party application I use Helm, and for simplicity I'm using Lens to interact with Helm instead of the command line.

The first thing I do is to install ingress-nginx to route custom URLs to the appropriate pods. Then, I install cert-manager to handle TLS certificate generation with Let's Encrypt.

Install ingress-nginx

From the Apps > Charts tab, I search ingress-nginx by ingress-nginx:

lens interface screenshot

Searching ingress-nginx in chart list

In the yaml configuration, I'm setting hostNetwork to true to bind ingress to 80 and 443 host's port then click Install:

lens interface screenshot

Installing ingress-nginx chart

This configuration is not recommended but it's an easy way to address the lack of a load balancer in front of the cluster.

Install cert-manager

From Lens, I go to Apps > Charts, and I search for cert-manager by Jetstack.

I select version 1.6.3 and click Install. It opens the yaml config, where I enable CRDs installation by switching installCRDs to true in the yaml config:

lens interface screenshot

Installing cert-manager chart

Once installation as finished, I run the following command from the k0s server to add a new certificate issuer:

cat <<EOF | sudo k0s kubectl apply -f -
kind: ClusterIssuer
name: letsencrypt
name: letsencrypt
- http01:
class: nginx

The email address I'm providing here will receive all the Let's Encrypt expiration notices. It can get annoying and for that reason I'm using a fake one.

Resource check

Note that 1 GB of RAM is already filled up, so it definitely takes at least a 2 GB.

Also 4 GB of disk space already filled up, so 16 GB is recommended:

$ df -h
Filesystem Size Used Avail Use% Mounted on
/dev/root 16G 3.9G 12G 25% /

An overview of how incoming traffic flows

Because I've set hostNetwork to true when installing ingress-nginx, it has created the following Kubernetes endpoint:

$ sudo k0s kubectl -n default get endpoints
ingress-nginx-...-controller, 2h

It allows the host's incoming HTTP and HTTPS traffic to be forwared to the cluster, and more specifically, to this pod (more on this in part IV):

$ kubectl get pod -A -l
default ingress-nginx-164... 1/1 Running 0 1d

The diagram below shows how incoming traffic flows throught components. Now that I've configured the cluster, so far I've set up the first three steps:

✓ 1.client DNS ok and 443/TCP port open
✓ k0s installed
✓ 3.ingress ingress-nginx installed

Next step

My cluster is now ready2 to host applications. In the next parts, I'll show how to automate the deployment of any branch or any commit of a repository from Gitlab CI/CD, generate a unique URL à la Vercel and promote any deployment to production.

A Vercel-like PaaS beyond Jamstack with Kubernetes and GitOps, part II: Gitlab pipeline and CI/CD configuration

1 I tried with a t2.micro with 1 GB of RAM which can be run for free as part of the AWS Free Tier offer and fits the minimal system requirements of k0s for a controller+worker node, but it ended up being pretty unstable.

2 Ready for testing, there's a lot to say about this setup but it's not meant to be a permanent solution. See in afterward.

About me


Hi, I'm Jonathan Experton.

I help companies start, plan, execute and deliver software development projects on time, on scope and on budget.

Montreal, Canada · GMT -4