Creating a Kubernetes Cluster

Kubernetes’ documentation describes the many ways to set up a cluster. Here, we shall provide quick instructions for the most painless and popular ways of getting setup in various cloud providers:


During the process of setting up JupyterHub, you’ll be creating some files for configuration purposes. It may be helpful to create a folder for your JuypterHub deployment to keep track of these files.


If you are concerned at all about security (you probably should be), see the Kubernetes best-practices guide for information about keeping your Kubernetes infrastruture secure.

Setting up Kubernetes on Google Cloud

Google Container Engine (confusingly abbreviated to GKE) is the simplest and most common way of setting up a Kubernetes Cluster. You may be able to receive free credits for trying it out. You will need to connect your credit card or other payment method to your google cloud account.

  1. Go to and log in.

  2. Enable the Container Engine API.

  3. Install and initialize the gcloud command-line tools. These tools send commands to Google Cloud and lets you do things like create and delete clusters.

    • Go to the gcloud downloads page to download and install the gcloud SDK.

    • See the gcloud documentation for more information on the gcloud SDK.

    • Install kubectl, which is a tool for controlling kubernetes. From the terminal, enter:

      gcloud components install kubectl
  4. Create a Kubernetes cluster on Google Cloud, by typing in the following command:

    gcloud container clusters create <YOUR_CLUSTER> \
        --num-nodes=3 \
        --machine-type=n1-standard-2 \


    • --num-nodes specifies how many computers to spin up. The higher the number, the greater the cost.
    • --machine-type specifies the amount of CPU and RAM in each node. There is a variety of types to choose from. Picking something appropriate here will have a large effect on how much you pay - smaller machines restrict the max amount of RAM each user can have access to but allow more fine-grained scaling, reducing cost. The default (n1-standard-2) has 2CPUs and 7.5G of RAM each, and might not be a good fit for all use cases!
    • --zone specifies which data center to use. Pick something that is not too far away from your users. You can find a list of them here.
  5. To test if your cluster is initialized, run:

    kubectl get node

    The response should list three running nodes.

Setting up Kubernetes on Microsoft Azure Container Service (ACS)


This is an alpha work-in-progress - please do not use in production! Help from people with more Azure experience would be highly welcome :)

  1. Install and initialize the Azure command-line tools, which send commands to Azure and let you do things like create and delete clusters.

  2. Authenticate the az tool so it may access your Azure account:

    az login
  3. Specify a Azure resource group, and create one if it doesn’t already exist:

    az group create --name=${RESOURCE_GROUP} --location=${LOCATION}


  • --name specifies your Azure resource group. If a group doesn’t exist, az will create it for you.
  • --location specifies which computer center to use. To reduce latency, choose a zone closest to whoever is sending the commands. View available zones via az account list-locations.
  1. Install kubectl, a tool for controlling Kubernetes:

    az acs kubernetes install-cli
  2. Create a Kubernetes cluster on Azure, by typing in the following commands:

    az acs create --orchestrator-type=kubernetes \
        --resource-group=${RESOURCE_GROUP} \
        --name=${CLUSTER_NAME} \
  3. Authenticate kubectl:

    az acs kubernetes get-credentials \
        --resource-group=${RESOURCE_GROUP} \


  • --resource-group specifies your Azure resource group.
  • --name is your ACS cluster name.
  • --dns-prefix is the domain name prefix for the cluster.
  1. To test if your cluster is initialized, run:

    kubectl get node

    The response should list three running nodes.

Setting up Kubernetes on Amazon Web Services (AWS)

AWS does not have native support for Kubernetes, however there are many organizations that have put together their own solutions and guides for setting up Kubernetes on AWS.

We like the Heptio guide, and recommend using this for setting up your cluster for clusters that span short periods of time (a week long workshop, for example). However, if you are setting up a cluster that would need to run for much longer, we recommend you use [kops]( It is a bit more complex, but provides features (such as log collection & cluster upgrades) that are necessary to run a longer term cluster.


The Heptio deployment of Kubernetes on AWS should not be considered production-ready. See the introduction in the Heptio Kubernetes tutorial for information about what to expect.

  1. Follow Step 1 of the Heptio guide, called Prepare your AWS Account.

    This sets up your Amazon account with the credentials needed to run Kubernetes.


    Make sure that you keep the file downloaded when you create the SSH key. This will be needed later to allow kubectl to interact with your Kubernetes cluster.


    You may find it helpful to “pin” the services we’ll be using to your AWS navbar. This makes it easier to navigate in subsequent sessions. Click the “pin” icon at the top, then drag CloudFormation and EC2 into your navbar.

  2. Deploy a Kubernetes template from Heptio.


    This section largely follows Step 2 of the Heptio guide.

    AWS makes it possible to deploy computational resources in a “stack” using templates. Heptio has put together a template for running Kubernetes on AWS. Click the button below to select the Heptio template, then follow the instructions below.

    You’ll be taken to an AWS page with a field already chosen under “Choose a template”. Simply hit “Next”.

    Enter AWS instance information (page 1): On this page you’ll tell AWS what kind of hardware you need. Fill in the following required fields:

    • Stack Name can be anything you like.
    • Availability Zone is related to the location of the AWS resources. Choose an AWS location close to your physical location or any other desired AWS location.
    • Admin Ingress Location defines the locations from which you can access this cluster as an administrator. Enter for the most permissive approach.
    • SSH Key is a dropdown list of keys attached to your account. The one you created in Step 1 should be listed here. This will allow you to SSH into the machines if you desire.
    • Node Capacity defines the number of machines you’ve got available. This will depend on the Instance Type that you choose. E.g., if you want each user to have 2GB and you expect 10 users, choose a combination of Instance Type and Node Capacity that meets this requirement.
    • Instance Type defines what kind of machine you’re requesting. See this list of instance types with Amazon as well as this list of pricing for each instance type.
    • Disk Size corresponds to the hard disk for each node. Note that this is different from the disks that users will use for their own notebooks/data. This disk should be large enough to contain the size of any Docker images you’re serving with the JupyterHub.
    • Instance Type (Bastion Host) corresponds to a computer that allows for easy SSH access to your Kubernetes cluster. This does not need to be a fancy computer. You may leave these as defaults. For more information on the Bastion Host, see here.

    Enter AWS instance information (page 2): On the second page you may leave all of these fields as is or customize as you wish. When done, hit Next. Then confirm and hit Next once more.

    AWS will now create the computational resources defined in the Heptio template (and according to the options that you chose).

    To see the status of the resources you’ve requested, see the CloudFormation page. You should see two stacks being created, each will have the name you’ve requested. When they’re done creating, continue with the guide.


    This often takes 15-20 minutes to finish. You’ll know it’s done when both stacks show the status CREATE_COMPLETE.

  3. Ensure that the latest version of kubectl is installed on your machine be following the install instructions.

  4. Configure your kubectl to send instructions to the newly-created Kubernetes cluster. To do this, you’ll need to copy a security file onto your computer. Heptio has pre-configured the command needed to do this. To access it, from the CloudFormation page click on the stack you just created (the one without “k8s-stack” in it). Below, there is an “Outputs” tab. Click on this, and look for a field called GetKubeConfigCommand. Copy / paste that text into your terminal, replacing the path/to/myKey.pem with the path to the key you downloaded in Step 1. It looks something like:

    SSH_KEY="<path/to/varMyKey.pem>"; scp -i $SSH_KEY -o
    ProxyCommand="ssh -i \"${SSH_KEY}\" ubuntu@<BastionHostPublicIP> nc
    %h %p" ubuntu@<MasterPrivateIP>:~/kubeconfig ./kubeconfig
  5. Tell Kubernetes to use this configuration file. Run:

    export KUBECONFIG=$(pwd)/kubeconfig
  6. Confirm that kubectl is connected to your Kubernetes cluster. Run:

    kubectl get nodes

    you should see a list of three nodes, each beginning with ip.

  7. Enable dynamic storage on your Kubernetes cluster. Create a file, storageclass.yml on your local computer, and enter this text:

    kind: StorageClass
      annotations: "true"
      name: gp2
      type: gp2

    Next, run this command:

    kubectl apply -f storageclass.yml

    This enables dynamic provisioning of disks, allowing us to automatically assign a disk per user when they log in to JupyterHub.

  8. Enable legacy authorization mode. This is temporarily required since the newer and more secure authorization mode is not out of beta yet.

    kubectl create clusterrolebinding permissive-binding \
     --clusterrole=cluster-admin \
     --user=admin \
     --user=kubelet \
This step should hopefully go away soon!

You should now be ready for the next step.

Next Step

Now that you have a Kubernetes cluster running, it is time to set up helm.