Google Cloud

In this tutorial, we'll create a Kubernetes v1.9.3 cluster on Google Compute Engine (not GKE).

We'll declare a Kubernetes cluster in Terraform using the Typhoon Terraform module. On apply, a network, firewall rules, managed instance groups of Kubernetes controllers and workers, network load balancers for controllers and workers, and health checks will be created.

Controllers and workers are provisioned to run a kubelet. A one-time bootkube bootstrap schedules an apiserver, scheduler, controller-manager, and kube-dns on controllers and runs kube-proxy and calico or flannel on each node. A generated kubeconfig provides kubectl access to the cluster.


  • Google Cloud Account and Service Account
  • Google Cloud DNS Zone (registered Domain Name or delegated subdomain)
  • Terraform v0.11.x and terraform-provider-ct installed locally

Terraform Setup

Install Terraform v0.11.x on your system.

$ terraform version
Terraform v0.11.1

Add the terraform-provider-ct plugin binary for your system.

tar xzf terraform-provider-ct-v0.2.0-linux-amd64.tar.gz
sudo mv terraform-provider-ct-v0.2.0-linux-amd64/terraform-provider-ct /usr/local/bin/

Add the plugin to your ~/.terraformrc.

providers {
  ct = "/usr/local/bin/terraform-provider-ct"

Read concepts to learn about Terraform, modules, and organizing resources. Change to your infrastructure repository (e.g. infra).

cd infra/clusters


Login to your Google Console API Manager and select a project, or signup if you don't have an account.

Select "Credentials", and create service account key credentials. Choose the "Compute Engine default service account" and save the JSON private key to a file that can be referenced in configs.

mv ~/Downloads/project-id-43048204.json ~/.config/google-cloud/terraform.json

Configure the Google Cloud provider to use your service account key, project-id, and region in a file.

provider "google" {
  version = "1.2"
  alias   = "default"

  credentials = "${file("~/.config/google-cloud/terraform.json")}"
  project     = "project-id"
  region      = "us-central1"

provider "local" {
  version = "~> 1.0"
  alias = "default"

provider "null" {
  version = "~> 1.0"
  alias = "default"

provider "template" {
  version = "~> 1.0"
  alias = "default"

provider "tls" {
  version = "~> 1.0"
  alias = "default"

Additional configuration options are described in the google provider docs.


A project may contain multiple clusters if you wish. Regions are listed in docs or with gcloud compute regions list.


Define a Kubernetes cluster using the module google-cloud/container-linux/kubernetes.

module "google-cloud-yavin" {
  source = "git::"

  providers = {
    google = "google.default"
    local = "local.default"
    null = "null.default"
    template = "template.default"
    tls = "tls.default"

  # Google Cloud
  region        = "us-central1"
  dns_zone      = ""
  dns_zone_name = "example-zone"
  os_image      = "coreos-stable-1576-5-0-v20180105"

  cluster_name       = "yavin"
  controller_count   = 1
  worker_count       = 2
  ssh_authorized_key = "ssh-rsa AAAAB3Nz..."

  # output assets dir
  asset_dir = "/home/user/.secrets/clusters/yavin"

Reference the variables docs or the source.


Initial bootstrapping requires bootkube.service be started on one controller node. Terraform uses ssh-agent to automate this step. Add your SSH private key to ssh-agent.

ssh-add ~/.ssh/id_rsa
ssh-add -L


terraform apply will hang connecting to a controller if ssh-agent does not contain the SSH key.


Initialize the config directory if this is the first use with Terraform.

terraform init

Get or update Terraform modules.

$ terraform get            # downloads missing modules
$ terraform get --update   # updates all modules
Get: git:: (update)
Get: git:: (update)

Plan the resources to be created.

$ terraform plan
Plan: 64 to add, 0 to change, 0 to destroy.

Apply the changes to create the cluster.

$ terraform apply Still creating... (10s elapsed)
... Still creating... (5m30s elapsed) Still creating... (5m40s elapsed) Creation complete (ID: 5768638456220583358)

Apply complete! Resources: 64 added, 0 changed, 0 destroyed.

In 4-8 minutes, the Kubernetes cluster will be ready.


Install kubectl on your system. Use the generated kubeconfig credentials to access the Kubernetes cluster and list nodes.

$ export KUBECONFIG=/home/user/.secrets/clusters/yavin/auth/kubeconfig
$ kubectl get nodes
NAME                                          STATUS   AGE    VERSION
yavin-controller-0.c.example-com.internal     Ready    6m     v1.9.3
yavin-worker-jrbf.c.example-com.internal      Ready    5m     v1.9.3
yavin-worker-mzdm.c.example-com.internal      Ready    5m     v1.9.3

List the pods.

$ kubectl get pods --all-namespaces
NAMESPACE     NAME                                      READY  STATUS    RESTARTS  AGE
kube-system   calico-node-1cs8z                         2/2    Running   0         6m
kube-system   calico-node-d1l5b                         2/2    Running   0         6m
kube-system   calico-node-sp9ps                         2/2    Running   0         6m
kube-system   kube-apiserver-zppls                      1/1    Running   0         6m
kube-system   kube-controller-manager-3271970485-gh9kt  1/1    Running   0         6m
kube-system   kube-controller-manager-3271970485-h90v8  1/1    Running   1         6m
kube-system   kube-dns-1187388186-zj5dl                 3/3    Running   0         6m
kube-system   kube-proxy-117v6                          1/1    Running   0         6m
kube-system   kube-proxy-9886n                          1/1    Running   0         6m
kube-system   kube-proxy-njn47                          1/1    Running   0         6m
kube-system   kube-scheduler-3895335239-5x87r           1/1    Running   0         6m
kube-system   kube-scheduler-3895335239-bzrrt           1/1    Running   1         6m
kube-system   pod-checkpointer-l6lrt                    1/1    Running   0         6m

Going Further

Learn about version pinning, maintenance, and addons.


On Container Linux clusters, install the container-linux-update-operator addon to coordinate reboots and drains when nodes auto-update. Otherwise, updates may not be applied until the next reboot.



Name Description Example
cluster_name Unique cluster name (prepended to dns_zone) "yavin"
region Google Cloud region "us-central1"
dns_zone Google Cloud DNS zone ""
dns_zone_name Google Cloud DNS zone name "example-zone"
ssh_authorized_key SSH public key for ~/.ssh_authorized_keys "ssh-rsa AAAAB3NZ..."
os_image OS image for compute instances "coreos-stable-1576-5-0-v20180105"
asset_dir Path to a directory where generated assets should be placed (contains secrets) "/home/user/.secrets/clusters/yavin"

Check the list of valid regions and list Container Linux images with gcloud compute images list | grep coreos.

DNS Zone

Clusters create a DNS A record ${cluster_name}.${dns_zone} to resolve a network load balancer backed by controller instances. This FQDN is used by workers and kubectl to access the apiserver. In this example, the cluster's apiserver would be accessible at

You'll need a registered domain name or subdomain registered in a Google Cloud DNS zone. You can set this up once and create many clusters with unique names.

resource "google_dns_managed_zone" "zone-for-clusters" {
  dns_name    = ""
  name        = "example-zone"
  description = "Production DNS zone"

If you have an existing domain name with a zone file elsewhere, just carve out a subdomain that can be managed on Google Cloud (e.g. and update nameservers.


Name Description Default Example
machine_type Machine type for compute instances "n1-standard-1" See below
controller_count Number of controllers (i.e. masters) 1 1
worker_count Number of workers 1 3
worker_preemptible If enabled, Compute Engine will terminate controllers randomly within 24 hours false true
networking Choice of networking provider "calico" "calico" or "flannel"
pod_cidr CIDR range to assign to Kubernetes pods "" ""
service_cidr CIDR range to assign to Kubernetes services "" ""
cluster_domain_suffix FQDN suffix for Kubernetes services answered by kube-dns. "cluster.local" ""

Check the list of valid machine types.


Set controller_count to 1. A bug in Google Cloud network load balancer health checking prevents multiple controllers from bootstrapping. There are workarounds, but they all involve tradeoffs we're uncomfortable recommending. See #54.


Add worker_preemeptible = "true" to allow worker nodes to be preempted at random, but pay significantly less. Clusters tolerate stopping instances fairly well (reschedules pods, but cannot drain) and preemption provides a nice reward for running fault-tolerant cluster systems.`