JupyterHub.nrw Platform Changelog¶

• Idle servers are now culled after 1 hour for students/members and 4 hours for employees, faculty and staff.
• Any running server is now force-stopped after 8 hours, regardless of activity.

Fix resource access policy resolution for identity providers that scope affiliations under a sub-domain, e.g. student@stud.hshl.de instead of student@hshl.de. These users previously matched no policy and received no resources. Also add student as a recognized affiliation alongside member.

The hub now restarts itself automatically if its Kubernetes watch on pods or virtualservices gives up (e.g. after apiserver throttling), instead of silently going blind to newly spawned pods. Previously this required a manual restart and caused spawns to fail with does not have a uid! until someone noticed.

• The spawn page now displays your university's name and includes it in the page title.

The log viewer linked from the spawn-pending page now auto-refreshes every 10 seconds instead of showing a static snapshot, and log-backend errors are now surfaced with clear status messages instead of failing silently.

• Fix incorrect redirect when trying to visit /hub/, /user/ and /services/ without being logged in.
• Enable the Start button on the spawn page only if the user provides an input.
• Enforce virtualservice recreation upon spawning a new user pod to avoid spawn failure.

• Upgrade JupyterHub to 5.5.0 via Zero-to-JupyterHub Helm chart 4.4.0.
• Increase the shared memory /dev/shm in each pod to be 512M instead of 64M.

Fix a bug with nslcd.conf in the uni-muenster context.

Make the v-prefixed tags appears first in the dropdown menu for each image on the spawn page

Renew TLS certificate for jupyterhub.nrw and all hub subdomains

• New NVIDIA L40S-12Q GPU profile on the spawn page, available to JupyterHub.nrw users.
• New ROS2-Gazebo image for JupyterHub.nrw, scoped to group o0drl25 at the University of Münster.
• Palma Storage is now available in JupyterHub.nrw, restricted to University of Münster users.

Upgraded JupyterHub to 5.4.4 via the Zero-to-JupyterHub Helm chart 4.3.3.

nodeSelector now correctly prevents CPU-only sessions from being scheduled on GPU nodes. Previously, such workloads could land on GPU hardware and waste capacity.