Closed consideRatio closed 1 year ago
Thanks, @consideRatio
Note that dropping testing against k8s 1.23 is motivated by wanting to make a new breaking release of action-k3s-helm that drops support for k8s 1.23 testing which is hard to maintain as well, see jupyterhub/action-k3s-helm#94.
👍🏾
I repeatedly failed to figure out why our test suite started failing in our oldest dependencies test using JupyterHub 1 in #725. I spend a lot of time looking into this as it was also a clue for test failures in binderhub that probably was unrelated it turns out. With this background of investing significant time into maintenance for backward compatibility, I suggested in #727 we drop support for the previous major JupyterHub versions in KubeSpawner 6.0.0.
In this PR I'm declaring it a requirement to use
jupyterhub>=4
directly even thoughjupyterhub>=2
practically works to my knowledge. I motivate going straight to>=4
to simplify future maintenance and reduce the amounts of tests we need to run in our test suite. My assumption is that there is little value to support jupyterhub 2, 3 and 4 in kubespawner 6+ released after jupyterhub 4, because the z2jh distribution and other users of kubespawner would upgrade to jupyterhub 4 at the same time they would update to a new version of kubespawner.In the spirit of reducing some maintenance load, I'd like to also stop testing against k8s 1.23 and let KubeSpawner 6 come without guarantees it will work, even though it currently is known to work.
I suggest KubeSpawner 6 and z2jh 3 both ship with a disclaimer of "k8s 1.23 is tested to work, but going onwards we won't test against k8s 1.23".
Note that dropping testing against k8s 1.23 is motivated by wanting to make a new breaking release of action-k3s-helm that drops support for k8s 1.23 testing which is hard to maintain as well, see https://github.com/jupyterhub/action-k3s-helm/issues/94.