opichals / ampernetacle

Ampernetacle

This is a Terraform configuration to deploy a Kubernetes cluster on Oracle Cloud Infrastructure. It creates a few virtual machines and uses kubeadm to install a Kubernetes control plane on the first machine, and join the other machines as worker nodes.

By default, it deploys a 4-node cluster using ARM machines. Each machine has 1 OCPU and 6 GB of RAM, which means that the cluster fits within Oracle's (pretty generous if you ask me) free tier.

It is not meant to run production workloads, but it's great if you want to learn Kubernetes with a "real" cluster (i.e. a cluster with multiple nodes) without breaking the bank, and if you want to develop or test applications on ARM.

Getting started

1. Create an Oracle Cloud Infrastructure account (just follow this link).
2. Have installed or install kubernetes.
3. Have installed or install terraform.
4. Have installed or install OCI CLI .
5. Configure OCI credentials.
6. Download this project and enter its folder.
7. terraform init
8. terraform apply

That's it!

At the end of the terraform apply, a kubeconfig file is generated in this directory. To use your new cluster, you can do:

Linux

export KUBECONFIG=$PWD/kubeconfig
kubectl get nodes

Windows

$env:KUBECONFIG="$pwd\kubeconfig"
kubectl get nodes

The command above should show you 4 nodes, named node1 to node4.

You can also log into the VMs. At the end of the Terraform output you should see a command that you can use to SSH into the first VM (just copy-paste the command).

Windows

It works with Windows 10/Powershell 5.1.

It may be necesssary to change the execution policy to unrestricted.

PowerShell ExecutionPolicy

Customization

Check variables.tf to see tweakable parameters. You can change the number of nodes, the size of the nodes, or switch to Intel/AMD instances if you'd like. Keep in mind that if you switch to Intel/AMD instances, you won't get advantage of the free tier.

Stopping the cluster

terraform destroy

Implementation details

This Terraform configuration:

• generates an OpenSSH keypair and a kubeadm token
• deploys 4 VMs using Ubuntu 20.04
• uses cloud-init to install and configure everything
• installs Docker and Kubernetes packages
• runs kubeadm init on the first VM
• runs kubeadm join on the other VMs
• installs the Weave CNI plugin
• transfers the kubeconfig file generated by kubeadm
• patches that file to use the public IP address of the machine

Caveats

This doesn't install the OCI cloud controller manager, which means that you cannot create services with type: LoadBalancer; or rather, if you create such services, their EXTERNAL-IP will remain <pending>.

To expose services, use NodePort.

Likewise, there is no ingress controller and no storage class.

These might be added in a later iteration of this project. Meanwhile, if you want to install it manually, you can check the OCI cloud controller manager github repository.

Remarks

Oracle Cloud also has a managed Kubernetes service called Container Engine for Kubernetes (or OKE). That service doesn't have the caveats mentioned above; however, it's not part of the free tier.

What does "Ampernetacle" mean?

It's a porte-manteau between Ampere, Kubernetes, and Oracle. It's probably not the best name in the world but it's the one we have! If you have an idea for a better name let us know. 😊

Possible errors and how to address them

Authentication problem

If you configured OCI authentication using a session token (with oci session authenticate), please note that this token is valid 1 hour by default. If you authenticate, then wait more than 1 hour, then try to terraform apply, you will get authentication errors.

Symptom

The following message:

 Error: 401-NotAuthenticated
│ Service: Identity Compartment
│ Error Message: The required information to complete authentication was not provided or was incorrect.
│ OPC request ID: [...]
│ Suggestion: Please retry or contact support for help with service: Identity Compartment

Solution

Authenticate or re-authenticate, for instance with oci session authenticate.

Capacity issue

Symptom

If you get a message like the following one:

Error: 500-InternalError
│ ...
│ Service: Core Instance
│ Error Message: Out of host capacity.

It means that there isn't enough servers available at the moment on OCI to create the cluster.

Solution

One solution is to switch to a different availability domain. This can be done by changing the availability_domain input variable. (Thanks @uknbr for the contribution!)

Note 1: some regions have only one availability domain. In that case you cannot change the availability domain.

Note 2: OCI accounts (especially free accounts) are tied to a single region, so if you get that problem and cannot change the availability domain, you can create another account.

Using the wrong region

Symptom

When doing terraform apply, you get this message:

oci_identity_compartment._: Creating...
╷
│ Error: 404-NotAuthorizedOrNotFound
│ Service: Identity Compartment
│ Error Message: Authorization failed or requested resource not found
│ OPC request ID: [...]
│ Suggestion: Either the resource has been deleted or service Identity Compartment need policy to access this resource. Policy reference: https://docs.oracle.com/en-us/iaas/Content/Identity/Reference/policyreference.htm
│
│
│   with oci_identity_compartment._,
│   on main.tf line 1, in resource "oci_identity_compartment" "_":
│    1: resource "oci_identity_compartment" "_" {
│
╵

Solution

Edit ~/.oci/config and change the region= line to put the correct region.

To know what's the correct region, you can try to log in to https://cloud.oracle.com/ with your account; after logging in, you should be redirected to an URL that looks like https://cloud.oracle.com/?region=us-ashburn-1 and in that example the region is us-ashburn-1.

Troubleshooting cluster creation

After the VMs are created, you can log into the VMs with the ubuntu user and the SSH key contained in the id_rsa file that was created by Terraform.

Then you can check the cloud init output file, e.g. like this:

tail -n 100 -f /var/log/cloud-init-output.log