Sprint 2 for Image Management Goal, Q3 2023

[yuvipanda]

There are a few sub goals to do this sprint. Specifics to be fleshed out over the next few days. This is currently only a draft

Measuring effects of prior interventions

Sprint 1: unlisted_choice on the LEAP hub

Let's do an assessment at the end of Sprint 2 to see how Sprint 1's intervention worked.

Tasks

Preview Give feedback

1. Measure use of unlisted_choice in LEAP hub to see if anyone is actually using it. #2875
   +0
   
2. Meet with the community champion for LEAP to better understand how they would actually use the unlisted_choice feature #2876
   +0
   

Intervention 2: profileList construction

Specifying changes to profileList is actually the biggest source of support tickets when we last tagged them, causing almost 3.5 times as many tickets as the next one. The next biggest one is actually about deploying a single image change, rather than something part of a profileList. Removing 2i2c engineers entirely from this would be very helpful. While we currently have a deployment of 'jupyterhub-configurator', we need to get rid of that and instead deploy the newer 'z2jh-configurator' so we have a path forward. See 2i2c-org/features#26 for more information here.

However, profileLists can be complex - particularly, our current "node share" setup is a bit complex and hard, and integrating that into a GUI will take time. Doing that along with getting a new configurator deployed will be too much work for one sprint.

Instead, we'll solve a slightly smaller problem that will help us along the way!

Problem to solve

Some hubs (Callysto & UToronto right now) use only one image per hub, but can not use the existing configurator because they also desire to use the image prePuller. This is currently possible only via editing config on GitHub. So you end up with PRs like #2823.

We can work with the Callysto team to test the following hypothesis:

If community champions can set the image to be used via a GUI, and it also configured pre-pulling appropriately, there would be no support tickets / GitHub PRs for changing the image tag.

This would help us with a few things:

1. Get z2jh-configurator deployed in our infrastructure in a sustainable way. This paves the way for more complete profileList functionality in the future.
2. Validate that we can get prePuller functionality integrated into z2jh-configurator - being able to provide functionality like this is one of the reasons to move to z2jh-configurator, and this lets us validate that.
3. No more support requests for image tag changes for hubs with just a single image in use.

However, we had initially scoped this goal to only focus on research hubs, and Callysto is not a research hub. It offers a progressive pathway to solving this for research hubs in a clean way that fits with our hypothesis driven spirnts though, so I think it is ok to try this out.

Sprint Plan

Tasks

Preview Give feedback

1. Talk to Callysto to make sure they're onboard with this
2. Make a helm chart for z2jh-configurator
3. Make a plan for how pre-puller would work with z2jh-configurator. Without this, z2jh-configurator would not be particularly helpful
4. Make a docker image for z2jh-configurator
5. Setup chartpress for z2jh-configurator

Dynamic image building

This is a longer term project for this goal to have a good demo by september.

Tasks

Preview Give feedback

1. Expand the hub-experimental image to accomodate multiple experiments #2883
   +0
   
2. Update the imagebuilding-demo hub use a hub-experimental image with the prototype kubespawner UI #2884
   +0
   

Prepare for future sprints

As we go along, we should work on building out future sprints.

Tasks

Preview Give feedback

1. Understand how we can best support Linux desktop on JupyterHub by talking to VICTOR folks #2862
   +0
   
2. Meet with openscapes folks to understand their use of profiles and node sharing #2882
   +0
   
3. Deploy unlisted_choice to researchdelight hub, so it acts as a demo #2873
   +0
   
4. Find 2 more communities that we think might benefit from unlisted_choice, demo the feature to them, and report back on how they would use it. Optionally, enable it for them as well. #2874
   +0
   

Carry-over from previous sprint

We will try to minimize carrying issues forward, but anything carried over should be listed here.

Tasks

Preview Give feedback

1. [Docs] Create documentation on how to choose a tag for pangeo-docker images #2793
   4 of 4
   +0
   
2. API Calls to spawn user servers do not handle defaults in profileList correctly jupyterhub/kubespawner#758
   bug +1

[damianavila]

More details about the evolution of the sprint live in the corresponding learning table and the retro notes.
Sprint 2 is finished, so closing this one.