Skip to content

Latest commit

 

History

History
341 lines (255 loc) · 13.2 KB

CONTRIBUTING.md

File metadata and controls

341 lines (255 loc) · 13.2 KB

Contributing

Welcome! As a Jupyter project, we follow the Jupyter contributor guide.

Setting up minikube for local development

We recommend using minikube for local development.

  1. Download & install minikube.

    For MacOS: You may install minikube using Homebrew brew cask install minikube or from a binary at https://github.com/kubernetes/minikube/releases. If you need to install Docker Community Edition (CE) for Mac, please follow the Docker instructions.

  2. Download & install helm.

    You may install Helm using one of the following steps:

    • With the following curl command:

      curl https://raw.githubusercontent.com/kubernetes/helm/master/scripts/get | bash
      
    • From one of the binaries at https://github.com/helm/helm/releases

    • For MacOS, using Homebrew: brew install kubernetes-helm

  3. Start minikube.

    For minikube version 0.26 and higher:

    minikube start

    For older minikube versions:

    minikube start --extra-config=apiserver.Authorization.Mode=RBAC

    Note on troubleshooting: if you recently upgraded minikube and are now seeing errors, you may need to clear out the ~/.minikube and ~/.kube directories and reboot.

  4. Use the docker daemon inside minikube for building:

    eval $(minikube docker-env)
  5. Clone the zero-to-jupyterhub repo:

    git clone [email protected]:jupyterhub/zero-to-jupyterhub-k8s.git
    cd zero-to-jupyterhub-k8s
  6. Create a virtualenv & install the libraries required for builds to happen:

    python3 -m venv .
    source bin/activate
    python3 -m pip install -r dev-requirements.txt
  7. Now run chartpress to build the requisite docker images inside minikube:

    chartpress

    This will build the docker images inside minikube & modify jupyterhub/values.yaml with the appropriate values to make the chart installable!

  8. Configure helm and minikube for RBAC:

    kubectl create clusterrolebinding add-on-cluster-admin --clusterrole=cluster-admin --serviceaccount=kube-system:default
    kubectl --namespace kube-system create sa tiller
    kubectl create clusterrolebinding tiller \
        --clusterrole cluster-admin \
        --serviceaccount=kube-system:tiller
    helm init --service-account tiller
  9. Install / Upgrade JupyterHub Chart!

    helm upgrade --wait --install --namespace=hub hub jupyterhub/ -f minikube-config.yaml

    You can easily change the options in minikube-config.yaml file to test what you want, or create another config.yaml file & pass that as an additional -f config.yaml file to the helm upgrade command.

  10. Retrieve the URL for your instance of JupyterHub:

minikube service --namespace=hub proxy-public

Navigate to the URL in your browser. You should now have JupyterHub running on minikube.

  1. Make the changes you want.

    To view your changes on the running development instance of JupyterHub:

    • Re-run step 7 if you changed anything under the images directory
    • Re-run step 9 if you changed things only under the jupyterhub directory.

Travis CI tests

Travis tests are automatically run on every pull request. Since the Travis environment is not accessible it can be difficult to debug CI failures. A Vagrantfile which partially simulates the Travis environment is included, and may be useful when updating the CI deployments, though it is by no means an exact replica.

  1. Start and login to the Vagrant box:

    cd ci
    vagrant up
    vagrant ssh
  2. Run the test script. Optionally edit SCENARIO in ./ci/vagrant-run.sh if you want to test a different scenario

    cd /zero-to-jupyterhub-k8s
    ./ci/vagrant-run.sh

Best practices

We strive to follow the guidelines provided by kubernetes/charts and the Helm Chart Best Practices Guide they refer to.

Releasing a new version of the helm chart

The following steps can be followed to release a new version of the Helm Chart. Presently, we expect a release approximately every 5-7 weeks.

Create an issue for the new release

Use this issue to coordinate efforts and keep track of progress. You can copy / paste the raw Markdown from the following list, which will be covered in more detail below.

Title: Release {{release-name}}
Content:

This issue will be used to coordinate the next release of the helm
chart, {{release-name}}. Instructions for creating the release can be found in
[CONTRIBUTING.md](https://github.com/jupyterhub/zero-to-jupyterhub-k8s/blob/master/CONTRIBUTING.md#releasing-a-new-version-of-the-helm-chart).
Below is the checklist for this release.

- [ ] Code, tests, and documentation to support a release are stable.
- [ ] Make a CHANGELOG
- [ ] Generate and add the list of contributors
- [ ] Build and push a new Docker image to DockerHub
- [ ] Commit version bump in `Chart.yaml` and `Values.yaml`
- [ ] Update references in documentation to the new version (note: documentation
      should be stable and there should be no anticipated major changes to content).
- [ ] Confirm that a new deployment using the updated instructions works
- [ ] Create and push a new tag for this release
- [ ] Create and publish a new GitHub release
- [ ] Write / publish a blog post based largely off of the CHANGELOG
- [ ] Set ReadTheDocs to begin using `latest` by default
- [ ] Celebrate!

As there are often many documentation improvements following the release of a new version, we set ReadTheDocs to serve latest/ until the first docs are written that are next-version-specific. As soon as documentation must be written for the next version of the Helm Chart, you can use the following checklist:

- [ ] Create a new tag for a documentation release (same release name with `-doc` at the end)
- [ ] Publish this tag
- [ ] Set ReadTheDocs to point to the **new tag** by default instead of `latest`
- [ ] Continue making next-version-specific changes to the documentation.

Note: Switching the documentation to latest after a new release is a stop-gap measure to accomodate the fact that the documentation is still changing relatively rapidly. Once the documentation as a whole stabilizes (after a few more release cycles), we plan to begin switching straight from the last version to the new version of documentation without going through latest.

Make a CHANGELOG

This needs to be manually created, following the format of current CHANGELOG. The general structure should be:

  • A short description of the general theme / points of interest for this release.
  • Breaking changes + a link to the upgrade instructions in the docs
  • A list of features with brief descriptions under each.
  • The contributor list mentioned in the section below.

Add list of contributors

We try to recognize all sorts of contributors, rather than just code committers.

Use the script in tools/contributors.py to list all contributions (anyone who made a commit or a comment) since the latest release. For each release, you'll need to find the versions of all repos involved:

Edit contributors.py to have the appropriate dates for each of these versions. Then, run the script and paste the output into the changelog. For an example, see the v0.5 list of contributors.

Push built images to DockerHub + bump version

The JupyterHub helm chart uses a Docker image that's registered on DockerHub. When releasing a new version of the helm chart, you also need to push a new version of this image. To do so, you must have:

  1. Docker running locally
  2. An account on DockerHub that you are logged into from your local docker installation.
  3. Push rights for images under jupyterhub/ on the DockerHub registry.
  4. Push rights to the jupyterhub/helm-chart repository on GitHub.
  5. A local SSH key that will let you push to the helm-chart repository on GitHub. See these instructions for information on how to create this.

Note: If you don't have a DockerHub account, or don't have push rights to the DockerHub registry, open an issue and ping one of the core devs.

If you have all of this, you can then:

  1. Check out latest master of z2jh
  2. Run chartpress --tag <VERSION> --push --publish-chart.
  • For example, to relase v0.5, you would run chartpress --tag v0.5 --push --publish-chart. Note the v before version.
  1. This will also modify the files Chart.yaml and values.yaml. Commit these changes.
  2. Look through the z2jh documentation and find any references to the Helm Chart version (e.g., look for the flag --version, as well as for all helm upgrade and helm install commands). Update these references to point to the new version you are releasing.
  3. Make a PR to the z2jh repository and notify the team to take a look.

After this PR gets merged:

  1. Go to https://zero-to-jupyterhub.readthedocs.io/en/latest and deploy a JupyterHub using the instructions (make sure that you're reading from /en/latest). Make sure your latest changes are present, and that the JupyterHub successfully deploys and functions properly.

Next, move on to making a GitHub release, described below.

Tagging and making a GitHub release

Now that our Docker image is pushed and we have updated the documentation for z2jh, it's time to make a new GitHub release. To do this, you must have:

  1. Push rights to the jupyterhub/zero-to-jupyterhub-k8s repo

You will need to make a git tag, and then create a GitHub release.

  1. Make sure you're on branch master with your latest changes from the section above pulled.
  2. Make a git tag with:
git tag -a <VERSION>

Where <VERSION> should be the new version that you're releasing. Note the v before the version number.

Git will ask you to include a message with the tag. Paste the entire contents of the CHANGELOG for this particular release. An easy way to do this is to paste the contents in a text file, and then refer to that text file with the call to commit: git tag -a <VERSION> -F <PATH-TO-FILE.txt> 3. Push the tags to the jupyterhub/zero-to-jupyterhub-k8s repo with git push <REMOTE-NAME> --tags. Note that <REMOTE-NAME> is whatever your local git uses to refer to the jupyerhub/ organization's repository (e.g., official or upstream) 3. Make a GitHub Release:

  • go to https://github.com/jupyterhub/zero-to-jupyterhub-k8s/releases and click 'Draft new release'.
  • The title should be the new version, followed by the name of the cricketer for the release. Like so:v0.5: "Hamid Hassan".
  • The description should include the entire changelog entry for this release.
  • Make sure the title/description/tag name look correct, and then click on Publish Release.

You've just made a GitHub release!

RTD update

Wait a few hours to let the release 'cool' and make sure that links, webpages, etc have updated. Then, update our documentation settings on readthedocs to show latest by default. This marks the official 'release' of the version!

Last step - release a blog post and tell the world!

The final step is to release a blog post. This doesn't have to be done by the person who performed all of the above actions.

To release a blog post for the new version, start a draft on the Jupyter Medium blog. Copy/paste the section of the CHANGELOG corresponding to the new release, then make minor modifications to make it more blog-friendly.

Don't forget to tell the JupyterHub community about the new release, and to encourage people to talk about it on social media!

That's it! Congratulations on making a new release of JupyterHub!

Extra step - release a documentation release

It is common that documentation changes are made shortly after a new release. To handle this, we often create a documentation release a few days after a major release.

To do this, confirm that all changes to the documentation are merged into master, then create a new tag with the same release name and -doc appended to the end. Create a GitHub release with the new tag and a description that points to the original release description. Finally, set our ReadTheDocs settings to point users to the new -doc tag by default instead of latest.