Deploying Pangeo on the Cloud

The commercial cloud is a powerful environment in which to deploy a Pangeo environment. However, cloud computing is unfamiliar to most researchers. The point of this guide is to describe how to setup up your own Pangeo cluster on the cloud platform of your choice, assuming zero knowledge of how the cloud works. We do assume you are comfortable using the unix command line and have access to one on your personal computer.


This guide, and the Pangeo cloud configuration in general, is based heavily on the Zero to Jupyterhub guide. We strongly recommend that your read that guide closely if you want a more comprehensive understanding.

This guide covers deployment on Google Cloud Platform. There are also deployment instructions available for the Azure cloud platform. We hope soon to expand this guide to also include deployment instructions for AWS.

Step One: Install the Necessary Software

Google Cloud SDK

The Google Cloud SDK is the command-line interface for Google Cloud Platform products and services. Download and installation instructions are at

Kubernetes kubectl

Kubernetes is an open source system for managing containerized applications in the cloud. The Kubernetes command-line tool, kubectl, allows you to deploy and manage applications on Kubernetes. It is necessary to deploy a Pangeo cluster.

If you have already installed the Google Cloud SDK, the easiest way to install kubectl is to run the following command on the command line:

gcloud components install kubectl

More comprehensive documentation for installing Kubernetes can be found at

Helm Client

Helm is a package manager for Kubernetes. Helm helps you automatically download and deploy the Pangeo default configuration for your cluster.

To install the helm command line client, follow the installation instructions at

To do this in one line from the command line, run:

curl | bash

Pangeo scripts template

Following chapters describe some batch scripts that are available in the pangeo repo. Clone the repo using git and go to the appropriate folder to use them

git clone
cd pangeo/gce/setup-guide

You can then edit your configuration parameters in file and then export them


Step Two: Create a Kubernetes Cluster

First, you must create an account on the cloud provider of your choice. Here we are using Google Cloud Platform.

All Google Cloud services are provisioned within the context of a project. If you haven’t done so already, you should create a project and switch to it in the Google Cloud console. From the APIs and Services section of the cloud console, you must then enable the Kubernetes Engine API for the project.

To create the cluster from the command line, you must first be authenticated. Run the following from the commmand line:

gcloud auth login

Following this step, you will need to run a set of bash commands to create a cluster with the components that Pangeo requires:

  • An incompressible default node pool for the Jupyterhub, web proxy, and user notebook servers.
  • An auto scaling node pool for Dask workers.

This script is availabe in pangeo/gce/setup-guide as, so you can use it directly.


If you expect your notebook images to be fairly large, it may be adventageous to setup your cluster to use faster SSD boot disks. This will typically provide faster boot times for notebooks and Dask workers. To do this, you’ll want to setup your cluster and any node pools with the --disk-type pd-ssd option. More information on how to configure SSD boot disks can be found in the GCP documentation.

Step Three: Configure Kubernetes

This script sets up the Kubernetes Role Based Access Control necessary for a secure cluster deployment.

This script is available in pangeo/gce/setup-guide as

Step Four: Create Cluster-Specific Configuration

There are two configuration files needed to deploy the Pangeo helm chart. Those files are available in the pangeo/gce/setup-guide folder of this repo. The first, jupyter_config.yaml, specifies modifications to the configuration that are unique to each deployment, so you will need to edit it.

The most important thing to configure here is the loadBalancerIP. If you have not reserved a static external IP, you can do so by running:

gcloud compute addresses create pangeo-jhubip --region $REGION
gcloud compute addresses list | grep pangeo-jhubip

The other file is secret_config.yaml, which specifies cluster specific encryption tokens.

The jupyterhub proxy secret token is a random hash, which you can generate as follows:

openssl rand -hex 32

Replace <SECRET> in the secretToken section with the output of this (note: you will stilll need the quotes around this value)

Most pangeo deployments use GitHub OAuth Callback, (or GitHub OAuth for developers) to authenticate users.

This authentication method needs an IP or domain name to work. This should be the IP you’ve reserved above, if you don’t have a domain name yet. Insert this IP in the following block:

      loadBalancerIP: <GCE_EXTERNAL_IP>


Replace clientId, clientSecret and callbackUrl values in the secret_config.yaml file with the values obtained from the GitHub app that you create for this.

Alternatively, you can also change authentication method, see the Zero to Jupyterhub guide for more information on that.

Step Five: Deploy Helm Chart

The following script deploys the most recent Pangeo Helm chart to your Kubernetes cluster.

If you want to use a specific version, check Pangeo Helm Chart for the version you want. You can then add a --version=0.1.1-a14d55b argument (for example) to helm install command, only keeping the last part of the release, without the pangeo-v prefix from the helm chart web-page.

This script is available as in the repo.

If you have not specified a static IP address in your configuration, the jupyterhub will come up at a random IP address. To get the address, run the command

kubectl --namespace=pangeo get svc proxy-public

Here’s what we see for when we run this commmand:

NAME           TYPE           CLUSTER-IP      EXTERNAL-IP    PORT(S)        AGE
proxy-public   LoadBalancer   80:30442/TCP   18d

The EXTERNAL-IP value is the address of the hub.

Upgrade Cluster

If you want to change the configuration, or to upgrade the cluster to a new version of the Helm Chart, run the following commmand (if you are just updating jupyterhub authentication IP, --force and --recreate-pods are not needed).

The scripts and are available for that.

Pangeo Helm Chart and Docker Images

Pangeo maintains its own Helm Chart and Docker images. These hold the default configuration for a Pangeo cloud deployment. These items live in the Pangeo helm-chart repository: