Automated deployment

[jtpio]

Currently:

• the gallery runs on a single machine using TLJH provisioned using this bootstrap command
• when the hooks have changed we need to run this command on the instance: sudo python3 the-littlest-jupyterhub/bootstrap/bootstrap.py --plugin tljh-voila-gallery/
• systemd units are periodically rebuilding the Docker images and resintalling the gallery package (Update the gallery package before building new images #50)

It should be possible to automate all of this while keeping the single instance setup:

• Setup CI for this repo (or a gallery-deploy repo)
• Trigger the deployment when merging into master, or using a simplified version of the mybinder.org recipe
• Use the openstack client with a voila-gallery-deploy user to start a new instance behind a floating IP (see Point voila-gallery.org to a floating IP #55) and do a rolling upgrade
• At the end of the TLJH setup, build the Docker images (only once)
• No systemd / scripts running periodically, so the instance stays "immutable"

[jtpio]

cc @yuvipanda

[jtpio]

Created https://github.com/voila-gallery/voila-gallery.org-deploy and opened voila-gallery/voila-gallery.org-deploy#1.

[jtpio]

There could be different options for building the Docker images:

1. trigger the build in the post_install hook. This will add a few minutes before the instance is ready
2. let CI build the images and push to DockerHub, and let the gallery pull them in the post_install hook (using latest or a specific tag from a ref field in gallery.yaml).

Option 1 might still be more adequate (at least for now), especially if we want to keep the gallery generic enough for those who want to deploy their own gallery and don't want to deal with container registries.

This again follows the idea of having an immutable TLJH instance with a specific set of examples. If the examples change they a new instance should be created.

[jtpio]

Just thinking out loud here.

Another option could be:

• only keep one instance alive
• when the gallery package is updated (code update or new example added), update the current instance
• ssh to the existing instance, update the gallery package and execute the bootstrap command (or with ansible)
• trigger the docker build only once
• restart JupyterHub to take the changes into account