The Elastisys Tech Blog


Kubernetes on GKE from scratch using Terraform


GKE, Google Container Engine, is an easy way of getting started with a managed Kubernetes service. In this post, we show how to get started from scratch and wind up with a cluster that is provisioned using Terraform. Defining your infrastructure in Terraform means that you can easily create a replica of your GKE environment. This is useful for e.g. a staging environment or to recover from a disaster. Let’s get started!

The goal of this blog post is to help you set up a cluster for serious testing or production use. If all you want to do is kick the tires a bit with Kubernetes, minikube is a better option. It runs on your machine and incurs no cloud costs. We will stand up the GKE environment using Terraform. This lets us manage its lifetime and create a new identcial one if needed. It also means that if disaster has struck, we do not need to hunt down the only person in the organization who knows how to deploy a full environment. Terraform encodes that knowledge.

Here’s a brief description of the cluster we will deploy:

  • It will have a single multi-purpose pool of nodes. If we need specialized ones later, we can add another node pool and set a Kubernetes Taint as appropriate.
  • The node pool should be managed entirely by GKE. This means it gets automatic upgrades of Kubernetes versions and automatic node repairs (or replacement).
  • The node pool should also have auto-scaling enabled, because we are not able to forecast how large it needs to be. And we do not want to turn customers away due to lack of resources, right?


There are a few things you should do before we get started for real:

Signing up for Google Cloud

This guide really is from scratch, and so, we do not even assume that you have an account at Google Cloud! If you do, then please jump to the next section.

Get your credit card ready and head over to Google Cloud. Hit the “Get started for free” button. While you get $300 of free credits, Google still needs a credit card to verify that you are not just some spam bot looking to score some free cloud resources. They won’t start charging you without telling you about it first, though.

Choose whether you want a business account or an individual one.

Signup complete at Google Cloud
Signup complete at Google Cloud

With that taken care of, we can start to create a project!

Creating a project

Google Cloud organizes all resources into folders and projects. It turns into a nice hierarchy like so:

Hierarchy of Google Cloud folders and projects
Hierarchy of Google Cloud folders and projects, from the official documentation

Terraform cannot help us set up this, so we will have to do this part manually using the Google Cloud console. The reason is that Terraform needs to work within the context of a project. So first create folders as appropriate for your organization’s needs, and then create a project.

New Google Cloud project created

Enabling the GKE API and creating a Service Account for Terraform

Terraform interacts with Google Cloud through its API (Application Programming Interface). And for security reasons, it first needs to be enabled for your project. So with your shiny new project defined, you can now enable it. Head over to Google Developers Console to enable the GKE API:

Enabling the GKE API

After hitting the button, you can access the Overview. It will tell you to create some credentials, and that is just what we will do.

For Terraform, we will want credentials of the Service Account type. Let’s head over to the IAM Admin console to do that. Hit Create Service Account and fill in the form like this:

Create Service Account dialog in Google Cloud IAM Admin console

Next step is to assign permissions. We want the “Kubernetes Engine Admin” and “Storage Admin” roles. The Kubernetes one is obvious, but the Storage one is used for Terraform itself. We will configure it to use Google Cloud Storage as the remote state backend later. This means that several operations people can access it later, without file inconsistencies. It should look like this:

Service Account permissions for Terraform

Finally, in the last step, the most important part is to create a key and downloading it in JSON format. Don’t miss this step!

Creating a key for the Service Account in JSON format

Save it to wherever you cloned the repo accompanying this blog post. Call it account.json, because that is what the Terraform files expect. It will not be committed to version control, so don’t worry. Also, keep it safe! Whoever has access to the key will be able to act as the Service Account. If you should lose it, disable the key immediately from the IAM Admin console.

Configuring Google Cloud SDK and Terraform

Not only do we want Terraform to use our Service Account, we want it to store its Remote State on Google Storage. This means that several team members can access the state. Because Terraform state is stored in the cloud, we can still manage our infrastructure even if the computer used for deploying it would go missing or similar.

Creating a bucket and enabling versioning

We will follow the official documentation on Google Storage for how to set up a bucket. Unfortunately, Google Storage console is currently unable to let us enable versioning, which is something Terraform strongly recommends (requires). For that, we need to download and install Google Cloud SDK, giving us access to the gsutil program. Once you have followed the instructions for the Google Cloud SDK, you can type the following command:

gcloud init

This will open a new browser window, asking you to log in. Do so, and answer all the questions in the terminal. Afterwards, you are ready to use the Google Cloud SDK.

The first thing we do with our Google Cloud SDK is enable versioning for our bucket. Type the following, exchanging BUCKET_ID with our value:

$ export BUCKET_ID=...your ID goes here...
$ gsutil versioning set on gs://${BUCKET_ID}
Enabling versioning for gs://gke-from-scratch-terraform-state/...
$ gsutil versioning get gs://${BUCKET_ID}
gs://gke-from-scratch-terraform-state: Enabled

If your output looks similar (apart from the bucket ID), you are good to go!

Configuring Google Cloud Storage as Terraform remote state

We will set up a Terraform Workspace for each of our deployments. Remember that there could be a reason for us to deploy several replicas of this environment (staging, disaster recovery, etc.). Let’s define one called “production”. There is a file in the accompanying repo called production.tfvars, which you can inspect, modify, and use.

Most values in the Terraform files have been written such that you just need to modify the production.tfvars file. However, one hard-coded value is in, on the line where the bucket is specified. The reason for this unfortunate limitation is technical: Terraform does not allow variables in that particular section. So change it in your file and then proceed.

To configure Terraform, we run:

$ export ENVIRONMENT=production
$ terraform workspace new ${ENVIRONMENT}
Created and switched to workspace "production"!

You're now on a new, empty workspace. Workspaces isolate their state,
so if you run "terraform plan" Terraform will not see any existing state
for this configuration.
$ terraform init -var-file=${ENVIRONMENT}.tfvars

Terraform gives us a lot of output, but among others, it will say that the “google” provider has been configured and that the “gcs” (remote state backend) provider has been set up. Success!

Using Terraform to create a GKE cluster

When you create a GKE cluster, you automatically wind up with an initial node pool. While that sounds nice, the problem is that we cannot manage it and set the configuration we want (auto-upgrades, auto-repair, and auto-scaling). So we will ask Terraform to create a cluster, then delete the initial node pool. It will then be asked to create a new node pool (that has the configuration options set), and then attach it to the cluster. This is so common (and the recommended approach) that Terraform has a special option for it.

Giving our Service Account required permissions

If we were to try to deploy now, we’d face a permissions error. We need to give our service account more permissions, and the roles/editor role. To do so, issue the following command:

$ export PROJECT=gke-from-scratch
$ export SERVICE_ACCOUNT=terraform
$ gcloud projects add-iam-policy-binding ${PROJECT} --member serviceAccount:${SERVICE_ACCOUNT}@${PROJECT} --role roles/editor

Deploying the GKE cluster

Now that we have done all the preparations, it is almost embarrassingly easy to deploy the cluster:

$ terraform apply -var-file=${ENVIRONMENT}.tfvars

Answer “yes” when prompted, and after a few minutes, you have a cluster! You can head over to the Kubernetes Clusters list in Google Cloud console and see it:

Kubernetes cluster listing in Google Cloud console


In this post, we showed how to deploy an auto-repairing, auto-updating, auto-scaling cluster from scratch on Google Container Engine (GKE) using Terraform. The setup made use of a custom Service Account for security, and used Terraform Workspaces to make multi-cluster management possible.

When your Kubernetes journey brings you to the point where a fully-managed and compliant Kubernetes cluster makes sense, don’t hesitate to contact us at Elastisys.