JupyterHub checks wrong server url after restart

Bug description

I use a subclass of BatchSpawner (created for Univa Grid Engine), JupyterHub 2.0.0. If I restart JupyterHub it uses wrong url to check if user’s server is still running. It concludes that server is not running and deletes cluster job.

Expected behaviour

Cluster jobs should survive JupyterHub restart

Actual behaviour

logs contain lines like these (note port 0) Verifying that tryauuum is running at http://my-hostname:0/user/tryauuum/ Of course server is not reachable via this url. After some time JupyterHub runs qdel to delete cluster job.

---

The bug is probably with my custom class

I see that after restart there still is a correct port and IP in sqlite database. I want to ask, where exactly in code server IP and port are read from servers table of sqlite database? do I have to implement it manually? All I see in code in terms of restoring IP and port from database is this iteration over list of spawners here https://github.com/jupyterhub/jupyterhub/blob/aa0ce1c88a8e34da1779cb13d3f36f81a6a24df7/jupyterhub/app.py#L2517-L2535

[minrk]

Should be fixed by https://github.com/jupyterhub/jupyterhub/pull/3859

[minrk]

I think you’re right, get_env probably shouldn’t modify the state. It also shouldn’t be called except during Spawner.start(), so I’m not entirely sure how/why this is coming up, but that much is easy to fix.