Tools used in this repo

Base tools

These are helpful for everyone, and not cloud provider-specific.

kubectl

The canonical commandline tool for talking to kubernetes clusters. Debugging and understanding runtime behavior of our hubs is greatly enhanced by knowledge of how to use kubectl. Understanding how to use kubectl really helps understand how kubernetes itself works.

Tips

  • Consider aliasing it to k, as you might be typing it a lot!

  • It is easy to accidentally operate on the wrong kubernetes cluster! Putting the current kubernetes context in your terminal prompt (via something like starship) can help a lot!

Additional resources and tools

  • The official cheatsheet, with extremely helpful tips.

  • Lens is an amazing GUI for exploring various objects in a kubernetes cluster. It can do almost everything kubectl does, in a fairly intuitive way. Highly recommended as a supplement (or even alternative) to kubectl.

  • k9s is a similar tool to Lens, but a terminal based UI.

  • Check out the list of kubectl plugins to see if anything will help with your workflow.

  • kubectx helps with switching between different k8s clusters and namespaces, which we might have to do often.

Helm

Helm is used in two ways:

  1. By our deployment scripts to deploy our hubs.

  2. To deploy cluster-wide support components (such as prometheus, grafana, nginx-ingress) for each cluster.

Tips

  • The --repo flag allow you to specify a Helm repo directly instead of needing to do helm repo add and helm repo update first.

    helm template --repo https://jupyterhub.github.io/helm-chart/ jupyterhub
    
  • The helm template command can quickly render templates for you which is a big part of developing a helm chart.

  • The helm template --show-only flag can help you review a specific template.

    # 1. An example on how to access a specific template
    helm template --repo https://jupyterhub.github.io/helm-chart/ jupyterhub \
        --show-only templates/hub/deployment.yaml
    
    # 2. An example on how to access a specific template from a subchart
    #
    #    Note we need to specify version as the helm repo doesn't contain a valid
    #    binderhub version that isn't considered a pre-release.
    helm template --repo https://jupyterhub.github.io/helm-chart/ binderhub \
        --version=0.2.0-n611.he777436 \
        --set jupyterhub.hub.services.binder.apiToken=asd \
        --show-only charts/jupyterhub/templates/hub/deployment.yaml
    
  • The helm template --validate flag can help you both render and then validate the rendered templates with a k8s api-server. This is an excellent lightweight test of a Helm chart in a CI system.

sops

In line with 2i2c’s Customer Right to Replicate, we try to keep all our deployment repositories as open as possible. But some values must be secret - like access tokens, cookie secret seeds, etc. We use sops to store them encrypted in our Git repo, so they can be version controlled and reviewed along with the rest of the repo. We use Google Cloud KMS to encrypt our secrets, so you need the Google Cloud tools installed and authenticated locally (following the instructions here) before you can use sops.

sops is called programatically by our deployment scripts to decrypt files for deployment, and you will use it interactively to modify or encrypt new files.

Terraform

Terraform is an “Infrastructure as Code” (IaC) tool that allows you to build, change, and version infrastructure in the cloud. We use terraform to provision cloud infrastructure and modify it directly. We then deploy applications on top of that infrastructure via Helm.

Google Cloud tools

google-cloud-sdk is the primary commandline tool used to interact with Google Cloud Platform (GCP). Our deployment scripts use it to authenticate to GCP, and it is very helpful in debugging node issues.

Tips

Authentication

gcloud has two authentication flows, and that can get quite confusing since we work on a number of clusters with different Google credentials.

gcloud auth login provides credentials for gcloud commands like gcloud compute instances list or gcloud container clusters list.

gcloud auth application-default login provides credentials for other tools (such as helm, kubectl, sops) to authenticate to Google Cloud Platform on your behalf. So if sops or kubectl is complaining about authentication, make sure you are authenticated correctly with application-default