We move this repositroy in maintenance mode for the following reasons:

• This was a MVP for an easy Kubernetes installation based on community tools like kubeadm and terraform
• The OpenStack integration is still brittle (e.g. SecurityGroups are not created in a reliable fasion)
• Critical features like updates are missing
• The setup is very opinionated (but simple)
• There are other tools to bootstrap a Kubernetes cluster on OpenStack (even if these tools are often more complex)

For an alternative take a look at kube-spray, kubeone or one of the other Kubernetes boostraping tools.

TLDR: This repositroy deploys an opinionated Kubernetes cluster on OpenStack with kubeadm and terraform.

After cloning or downloading the repository, follow these steps to get your cluster up and running.

Take a look at the example provided in the example folder. It contains three files: main.tf, provider.tf, and variables.tf. Have a look at main.tf. Customize settings like master_data_volume_size or node_data_volume_size to your needs, you might have to stay below quotas set by your OpenStack admin. Pick an instance flavor that has at least two vCPUs, otherwise kubeadm will fail during its pre-flight check.

We assume example to be your working directory for all following commands.

The Kubernetes cluster will use Keystone authentication (over a WebHook). For mor details look at the official docs or just use the quick start:

VERSION=1.13.1
OS=$(uname | tr '[:upper:]' '[:lower:]')
curl -sLO "https://github.com/kubernetes/cloud-provider-openstack/releases/download/${VERSION}/cloud-provider-openstack-${VERSION}-${OS}-amd64.tar.gz"
tar xfz cloud-provider-openstack-${VERSION}-${OS}-amd64.tar.gz
rm cloud-provider-openstack-${VERSION}-${OS}-amd64.tar.gz

mkdir $(pwd)/bin
cp ${OS}-amd64/client-keystone-auth $(pwd)/bin/
rm -rf ${OS}-amd64

As long as you keep the example folder inside the module repository, the reference source = "../" in the main.tf works. For a cleaner setup, you can also extract the example folder and put it somewhere else, just make sure you change the source setting accordingly. You can also reference the GitHub repository itself like so:

   module "my_cluster" {
     source = "git::https://github.com/inovex/kubernetes-on-openstack.git?ref=v1.0.0"

     # ...
   }

If you do it that way, make sure to

terraform get --update

before running any other terraform commands.

There a multiple different ways to authenticate with your OpenStack provider that all have their pros and cons. If you want to know more, check out this blog post about OpenStack credential handling for terraform. You can choose any of them, as long as you make sure the terraform variables auth_url, username and password are set explicitly as terraform variables. This is required as they are passed down to the Openstack Cloud Controller running inside the provisioned Kubernetes. Those should be dedicated service account credentials in a team setup. The easiest way to get started is to create a terraform.tfvars file in the example folder. If you start working in a team setup, you might want to check out the method using clouds-public.yaml, clouds.yaml and secure.yaml files in the aforementioned blog post.

Initialize the folder and run plan:

terraform init
terraform plan

Now you can create the cluster by running

terraform apply

It takes some time for the nodes to be fully configured. After running terraform apply there will be a kubeconfig file configured for the newly created cluster. The --insecure-skip-tls-verify=true in there is needed because we use the auto-generated certificates of kubeadm. There are possible workarounds to remove the flag (e.g. fetch the CA from the Kubernetes master, see below). Keep in mind: As a default all users in the (OpenStack) project will have cluster-admin rights. You can access the cluster via

kubectl --kubeconfig kubeconfig get nodes

It is also possible to set the KUBECONFIG environment variable to reference the location of the kubeconfig file created by terraform or to copy its contents to your .kube settings but keep in mind that the kubeconfig changes often because of Floating IPs.

To create a simple deployment, run

kubectl --kubeconfig kubeconfig create deployment nginx --image=nginx
kubectl --kubeconfig kubeconfig expose deployment nginx --port=80

In the current setup the master node can be reached by

ssh ubuntu@<ip>

and can also be used as jumphost to access the worker nodes:

ssh -J ubuntu@<ip> ubuntu@node-0

In order to prevent to use insecure-skip-tls-verify=true you can fetch the cluster CA:

export MASTER_IP=""
export CLUSTER_CA=$(curl -sk "https://${MASTER_IP}:6443/api/v1/namespaces/kube-public/configmaps/cluster-info" | jq -r '.data.kubeconfig' | grep -o 'certificate-authority-data:.*' | awk '{print $2}')
# ${CLUSTER_NAME} must match the name provided in the terraform.tfvars
export CLUSTER_NAME=""

kubectl --kubeconfig ./kubeconfig config set clusters.${CLUSTER_NAME}.certificate-authority-data ${CLUSTER_CA}
kubectl --kubeconfig ./kubeconfig config set clusters.${CLUSTER_NAME}.insecure-skip-tls-verify false

unset CLUSTER_CA
unset MASTER_IP
unset CLUSTER_NAME

• cloud-provider-openstack
• calico

In order to create a shared Kubernetes cluster for multiple users we can use application credentials

openstack --os-cloud <cloud> --os-project-id=<project-id> application credential create --restricted kubernetes

more docs will follow when the feature is merged.

If you want to use containerd in version 1.2.2 you will probably face this containerd issue if you use images from quay.io