renku
renku copied to clipboard
SwissDataScienceCenter
Provide user feedback when JupyterHub terminates a pod
Description
JupyterHub 1.0 will terminate pods that take longer than a configurable timeout to start. If this happens, we should inform the user about it, not just silently show that there are no environments starting anymore.
@ciyer the timeout will probably be exceeded if there is a problem starting the pod (like a missing image) - in that case I guess we should show the error straight away.
@olevski is it possible for the notebook-service to retrieve information about this either from k8s or from jh? If so, then we should return this info in in the response to the GET on a particular server.
I am sorry for being late to this. I will do some digging to see how the JH API reports a launch of a notebook that has timed out (i.e. which status code and if any message is provided).
Also I think the proper place to report this information is not by GET on servers but in the response to the POST request that is attempting to launch the notebook that timed out.
Also @ciyer do you know which conditions trigger the behaviour you mention above? And can you post them? It would be helpful to recreate it and see what is happening in the notebooks api and in the JH api.
you can make an image that's missing the jupyterhub-singleuser script so the container will fail to start - it will briefly say it's running, then change to pending and eventually disappear.