This repository manages the continuous deployment of the Pangeo Cloud Federation JupyterHub Kubernetes clusters using hubploy. It contains scripts to automatically redeploy when the image definition or chart parameters are changed. Changing the image will typically take ~20 minutes, and changing a Helm config variable ~1 minute.

Name	Cloud: region	Staging	Production
gcp-uscentral1b	GCP: us-central1-b	https://staging.us-central1-b.gcp.pangeo.io	https://us-central1-b.gcp.pangeo.io
aws2	AWS: us-west-2	https://staging.aws-uswest2.pangeo.io	https://aws-uswest2.pangeo.io
ooi	Azure: eastus	https://staging.ooi.pangeo.io	https://ooi.pangeo.io

The first step to using this automation is to create a Kubernetes cluster. Scripts to do so using Google Cloud Platform can be found here. For other cloud providers (e.g. AWS, Azure), follow the Zero-to-JupyterHub guide.

You will need to install git-crypt. git-crypt is used to encrypt the secrets that are used for deploying your cluster. Please read this HOW GIT-CRYPT WORKS if new to it.

Once you have a cluster created, you can begin customizing the configuration.

• Create a fork of this repository in GitHub and clone your fork. (Note: the default branch is staging.)
• Request a git-crypt symmetric key from the maintainers of this repo to be used for your deployment secrets files.
• Initialize git-crypt using the unlock command.
  • git-crypt unlock /path/to/your.key
• Copy the one of the deployments to a directory with your deployment name (we'll use foobar as our deployment name from here on).
  • cp -r example foobar
• Add your deployment's secrets directory to .gitattributes. IMPORTANT: before pushing to GitHub ensure encryption with git-crypt status | grep secrets
• Configure the JupyterHub config files. These are found in deployments/foobar/config.
• Configure the hubploy.yaml config file.
• Configure the deployment secrets found in deployments/foobar/config.
  • Information on what needs to be in hubploy.yaml and in the secrets directory can be found here.
• Configure your deployments image. This is found in deployments/foobar/image.
  • Edit the files in the binder directory to change the contents of the user Docker image. The specification for these files comes from repo2docker.
  • Add or modify the README.md and Jupyter notebooks. These will be in each user's home directory.

• Error: UPGRADE FAILED: "example.pangeo.io-staging" has no deployed releases
  • If your first deploy of an application fails. Run helm delete example-staging --purge anywhere you have run gcloud container clusters get-credentials

We have some rudimentary testing of deployments. The tests are located in the file test.py. Tests should be grouped according to which cloud deployment they should run on (all, GCP only, etc.)

There were some manual setup steps

• Sign up for the cloud with a regular user (we're using pangeo-bot)
• Create an API token, upload to CircleCI

We've deployed prometheus and grafana to monitor cluster usage. These metrics are publically visible:

• GCP: http://grafana.us-central1-b.gcp.pangeo.io/grafana/
• AWS: http://grafana.aws-uswest2.pangeo.io/grafana/

These are deployed independently of CI. The GCP deployment is done in deployments/gcp-uscentral1b/Makefile, with the metrics target.

$ cd deployments/gcp-uscentral1b
$ make metrics

Configuration is in the metrics folder. We expose it with an nginx-ingress, which was configured to have a static IP on Google Cloud and added to our DNS to server from the URL above.

Likewise with the aws / icesat2 deployment.

All pull requests should be made to the staging branch. When a change has been verified and is ready for deployment, staging can be merged into prod. This repository is set up with continuous deployment upon pushes to both staging and prod.

• Pangeo: main website for the Pangeo project.
• Pangeo Helm Chart: A simple helm chart that wraps the jupyterhub helm chart to support horizontal compute scaling with Dask.
• Zero to JupyterHub: A tutorial to help install and manage JupyterHub on a cloud with Kubernetes.
• HubPloy: a suite of commandline tools & a python library for continuous deployment of JupyterHub on Kubernetes (with Zero to JupyterHub)
• Repo2Docker: a tool to build, run, and push Docker images from source code repositories that run via a Jupyter server.