Skip to content

Latest commit

 

History

History
95 lines (64 loc) · 2.45 KB

README.md

File metadata and controls

95 lines (64 loc) · 2.45 KB

Deploying GFTS Hub

This is a log and record of deploying JupyterHub for GFTS

As much as possible, deployment uses automation via OpenTofu, helm, but there are always some manual steps.

Initial manual steps:

  1. create bucket for storing tofu state. Create user and store in secrets/ovh-creds.sh, and put bucket name in s3 backend configuration
  2. create API token for OVH API, store in secrets/ovh-creds.sh

Next, run tofu within the /deploy/tf directory, which will create the kubernetes cluster

tofu init
tofu plan
tofu apply

At this point, we have a kubernetes cluster. Export the kubernetes cluster config:

export KUBECONFIG=$PWD/../jupyterhub/secrets/kubeconfig.yaml
tofu output -raw kubeconfig > $KUBECONFIG
chmod 600 $KUBECONFIG
kubectl config rename-context kubernetes-admin@gfts gfts
kubectl config use-context gfts

And login to the private image registry:

echo $(tofu output -raw registry_builder_token) | docker login $(tofu output -raw registry_url) --username $(tofu output -raw registry_builder_name) --password-stdin

Now we move to the jupyterhub directory.

Build the image with chartpress:

chartpress --push

and deploy the chart with:

python deploy.py

Now jupyterhub should be running at https://gfts.minrk.net

Background

tofu is used to deploy cloud resources. Its configuration is in the jupyterhub/tofu directory. We only need to use tofu Once we have kubernetes running, we don't use tofu much anymore. tofu is not run on CI, because its actions can be quite destructive.

helm is used to deploy things on kubernetes. This is the main mechanism by which we modify our jupyterhub deployment. This can be done on CI (but isn't yet).

There are two configuration files:

  • gfts-hub/values.yaml is the main configuration file
  • secrets/config.yaml is the file containing

chartpress is used to build our user image and update the helm chart

Deploying updates is two steps:

  1. chartpress to ensure the image is up-to-date
  2. helm upgrade to apply the changes

To deploy an update:

python3 deploy.py chartpress
python3 deploy.py helm

and cleanup your local files:

chartpress --reset

The user image

The user image is defined in images/user. To change what's in the image, modify images/user/environment.yml and run make images/user/conda-linux-64.lock.