Gate Kernels on Install Status

[pmeier]

Summary

When the nebi server adopts the explicit-install model (see nebari-dev/nebi#438), workspaces default to locked-but-not-installed. nb_nebi_kernels must stop discovering kernels from uninstalled workspaces. One change: pass --installed to the local nebi workspace list call. No other touchpoints in this chart are affected.

Goals

• nb_nebi_kernels only discovers kernels from workspaces that are installed on disk (.pixi/envs/ present).
• The nb_nebi_kernels version pin in the jupyterlab image is bumped to the release containing the fix.

Non-Goals

• Changes to the jhub-apps environment selector (03-nebi-envs.py). It lists workspaces from the nebi server API for use in app pods. App pods always install into a fresh directory regardless of the server's install status, so filtering is unnecessary.
• Changes to the nebi serve configuration (images/nebi/jupyter_server_config.py). The nebi binary handles install status internally.
• Changes to workspace PVC setup. The binary discovers install status from disk via the existing PVC mount.

Background / Motivation

The nebi server is changing from implicit install (run pixi install on every workspace creation/update) to explicit install (pixi lock only; install is a user action). After this change:

• nebi workspace list --json returns all workspaces regardless of install status. A --installed flag filters to installed-only.
• Workspaces without .pixi/envs/ are locked but not runnable.

nb_nebi_kernels calls nebi workspace list --json without --installed, creating kernel specs for every locally-tracked workspace. Launching a kernel against an uninstalled workspace would trigger an implicit pixi install with no progress indication — the exact UX problem the nebi change was designed to eliminate.

The main JupyterLab pod gets workspaces through the persistent workspace PVC at /var/lib/nebi/workspaces. nb_nebi_kernels discovers kernels from these local workspaces and must only surface installed ones.

Design

1. Kernel discovery: pass --installed to nebi workspace list

Repo: nebari-dev/nb-nebi-kernels
File: src/nb_nebi_kernels/discovery.py

Change the subprocess call in discover_workspaces():

# Before
["nebi", "workspace", "list", "--json"]

# After
["nebi", "workspace", "list", "--json", "--installed"]

No other changes to nb_nebi_kernels. The --installed flag is part of the nebi CLI change in the same nebi release.

2. Version bump

File: images/jupyterlab/pixi.toml

Bump nb_nebi_kernels from ==0.1 to the release containing the --installed fix (e.g., ==0.2). The exact version depends on the release tag chosen for the nb_nebi_kernels change.

Tradeoffs & Risks

Tradeoff	Mitigation
The local nebi binary is part of this chart (copied via init container from nebi.image). The nebi workspace list --installed flag must exist in that image.	The nebi image tag is pinned in values.yaml. When bumped to include the --installed feature, the local CLI and nb_nebi_kernels change ship atomically.
The nb_nebi_kernels change is in a separate repo, creating a two-repo release dependency.	Both repos are under project control. The nb_nebi_kernels release must land before the data-science-pack version bump.