The biggest problem here is that our reflector only watches one namespace. Should we have multiple watchers? Multiple kinds of watchers?

Perhaps we need a kind of SharedInformer design that lets us hold multiple watchers? Don't want to have one watcher per pod.

Is anyone working on this currently? It would be a great feature to have...

I will be also interested in having this feature.

I'm looking to integrate JupyterHub into an existing k8s cluster which utilizes a large number (>1000) of per-user namespaces.

As discussed in #79 , templating Namespace seems like a reasonable approach. Our namespaces are provisioned externally; would anyone object to deferring ns creation/cleanup to a future enhancement?

Based on @yuvipanda's comment, I think those would actually be separate enhancements anyways @agt-ucsd? I think this PR handles the creation/cleanup and then we need to develop a separate one which handles the watching multiple namespaces which is the thing you want?

@zflamig thanks, it sounds like you're in agreement that creation/cleanup of per-user namespaces is separable from generic support for multiple user namespaces (which besides converting namespace to a template/callable, would also entail changes to watching). I'll open a separate issue for the latter.

@yuvipanda
Hi all, How is this feature going?

I believe this has been fixed by #458