Skip to main content

jupyterlab-rtclauncher

A small JupyterLab 4 extension for the NASA Research Platform. It adds Launcher tiles — under a Collaboration section — that start, join, and stop the project's shared real-time-collaboration (RTC) server.

Without it, members reach the shared server through JupyterHub's admin UI (/hub/admin), which is why they were granted the broad-looking admin-ui scope. This extension replaces that route so the scope can be dropped.

How it works

The tiles are state-driven: the Launcher shows only the actions that are valid right now, and repaints itself as the shared server's state changes.

server down  │ [Start abtesting Collaboration]        ──▶ new tab: /hub/spawn/abtesting-collab
─────────────┼──────────────────────────────────────────────────────────────────────────────
spawning     │ [Starting abtesting Collaboration…]    ──▶ new tab: hub progress page
─────────────┼──────────────────────────────────────────────────────────────────────────────
server up    │ [Join abtesting Collaboration]         ──▶ new tab: /user/abtesting-collab/lab
             │ [Stop abtesting Collaboration]         ──▶ confirm ──▶ DELETE server
─────────────┼──────────────────────────────────────────────────────────────────────────────
stopping     │ [Stopping abtesting Collaboration…]    ──▶ (re-checks status)
  1. On activation the frontend calls GET /jupyterlab-rtclauncher/config, which reads the spawn-time environment and answers with the project and shared account name.
  2. It then polls GET /jupyterlab-rtclauncher/status to track whether the shared server is running — every 15s when settled, every 3s while a spawn or stop is in flight, and immediately whenever the tab regains focus.
  3. Start opens <hubPrefix>/spawn/<project>-collab, Join opens /user/<project>-collab/lab, both in a new browser tab. Stop posts to /jupyterlab-rtclauncher/stop after a confirmation dialog, which proxies DELETE /hub/api/users/<project>-collab/server.

All three commands also live permanently in the command palette, where they dim via isEnabled instead of appearing and disappearing.

Why separate tiles, and why they are swapped rather than relabelled

A single toggling tile hid Join entirely — a second member had no way in except to stop the session — and it put a destructive action on the same pixel as a benign one.

Three permanently-visible tiles were the alternative, but a Launcher tile has no disabled state to fall back on: Card in @jupyterlab/launcher never consults commands.isEnabled, so a "greyed out" tile is still fully clickable. Showing only the valid subset is the only honest option there.

That subset is applied by adding and disposing launcher items, not by changing a command's label. app.commands.notifyCommandChanged() does nothing for the Launcher: it is a VDomRenderer bound to LauncherModel.stateChanged and ignores the command registry's commandChanged signal, so a state-dependent label renders once and then goes stale until a browser reload. launcher.add() and disposing its return value are what emit stateChanged.

Why a new browser tab

The shared server is a different JupyterHub user server on its own URL prefix, so entering it is a full-page navigation. Done in the current tab it evicts the user's own running notebook from their only window. window.open must be called synchronously from the click handler or pop-up blockers kill it; a blocked pop-up is reported with the address so the user can open it manually.

Why stopping matters

The shared server bills the project pool for as long as it runs. JupyterHub's idle culler reaps it after 1h idle, but an open browser tab keeps refreshing last_activity and can defeat that — and since the admin-ui scope was dropped, members have no other way to shut it down. The Stop action closes that gap.

Note the server's 10Gi EBS home persists whether it runs or not; stopping releases the compute, which is the expensive part.

Why the stop goes through the server extension

The hub's _xsrf cookie is scoped to Path=/hub/, so JavaScript served from /user/<name>/lab cannot read it — and JupyterHub 5 requires that XSRF token for cookie-authenticated, non-GET API calls. The stop therefore proxies through this extension's server handler, which authenticates with the notebook server's own JUPYTERHUB_API_TOKEN.

That token only carries admin:servers!user=<project>-collab when the user genuinely holds it: JupyterHub resolves Spawner.server_token_scopes against the user's own scopes and drops the rest. A user outside the project's Keycloak group gets a token without it and the hub answers 403. The extension grants nothing.

Requires rp-deployment to request that scope via Spawner.server_token_scopes (terraform/config/20-nojovyan.py). Without it, can_control is false and the tile degrades to start-only.

Which projects get a tile

Any project with collaboration enabled — there is no hardcoded project list and no allowlist to configure here.

The enablement signal is the RP_COLLAB_ACCOUNT environment variable:

  1. A platform admin flips Real-Time Collaboration → Enable collaboration for a project in the RPAdmin console, which provisions its login-less <key>-collab principal.
  2. RPAdmin's GET /api/me/spawn-identity — already called on every spawn — now returns collab_account: "<key>-collab" for that project, or null.
  3. rp-deployment terraform/config/20-nojovyan.py forwards a non-null value to the notebook as RP_COLLAB_ACCOUNT.
  4. This extension shows the working tiles only when that variable is present.

So enabling a new project is a toggle in RPAdmin — no change to this extension, the spawner, or the image. OSSProject supplies the project key for the tile's label; the account name comes from RP_COLLAB_ACCOUNT so the naming convention lives only in RPAdmin.

Projects without collaboration still get a tile (0.4.0) — a dimmed, informational one that says so when clicked, naming the Project Administrator as who can enable it. Before that, such a project simply had no tile, so nobody in it could discover the feature existed. The dimmed tile grants nothing and calls nothing; its only action is opening a dialog. It is still absent inside the shared collaboration server itself, where there is nothing left to launch.

Operational note. Since rp-deployment feature 100, RPAdmin's toggle is the only gate: the hub reads the enabled projects from /api/internal/collab-projects at startup and builds its JupyterHub roles from that, so this extension's tile and the hub's authorization can no longer disagree. What a newly enabled project still needs is a hub restart, since JupyterHub roles are startup config; until then the working tiles appear and the spawn is refused. Before feature 100 this was a second, hand-maintained RP_COLLAB_PROJECT_KEYS allowlist in terraform, which is now only the boot-time fallback for an unreachable RPAdmin.

The tile grants nothing

It is a link. Starting the shared server requires admin:servers!user=<acct> and entering it requires access:servers!user=<acct>; both are granted to the project's Keycloak /smdc-oss/<project> group by rp-deployment terraform/config/41-collab-accounts.py, and auto-revoke when Keycloak drops the membership. A user without them who follows the link is refused by the hub.

The working tiles are replaced by the dimmed, informational one when RP_COLLAB_ACCOUNT is unset (collaboration is off for the project), and the Launcher shows nothing at all when the current server is the collaboration server (JUPYTERHUB_USER ends in -collab). The server extension reports which of those two it is as unavailable_reason; they were a single state until 0.4.0, which is why the off-for-this-project case used to be invisible.

Install

pip install jupyterlab-rtclauncher

Both the frontend and the server extension are enabled automatically. To verify:

jupyter labextension list   # jupyterlab-rtclauncher ... enabled OK
jupyter server extension list

Development

pip install -e ".[test]"
jupyter labextension develop . --overwrite
jlpm build

Run the tests:

python -m pytest tests/   # server extension
jlpm test                 # frontend

Rebuild after a source change with jlpm build, or run jlpm watch in one terminal and jupyter lab in another.

Uninstall

pip uninstall jupyterlab-rtclauncher
  • Feature 083 — RTC collaboration accounts (rpadmin/specs/083-rtc-collab-accounts/)
  • jupyterlab-resource-tracker — the usage & cost extension shipped in the same image

Metadata

Release files for jupyterlab-rtclauncher 0.4.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for jupyterlab-rtclauncher 0.4.0
File Size Uploaded Provenance
jupyterlab_rtclauncher-0.4.0.tar.gz 160.6 kB Without provenance Details

Built distribution (wheel)

Table of built distributions (wheels) for jupyterlab-rtclauncher 0.4.0
File Interpreter ABI Platform
jupyterlab_rtclauncher-0.4.0-py3-none-any.whl Python 3 none any Details

Total release size: 204.6 kB

Release files / jupyterlab_rtclauncher-0.4.0.tar.gz

Download URL jupyterlab_rtclauncher-0.4.0.tar.gz
Size 160.6 kB
Tags Source
SHA-256 checksum
How to use checksums
ecf37052d07e58dce27b6381a8160b0dff97fec41afdb9b96feeea3542c4a581
BLAKE2b-256 checksum
How to use checksums
3489130a7dc2da59ecc6028fdd060e9c267ea3323accb2465d5565955ec88cac
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.14.6

Release files / jupyterlab_rtclauncher-0.4.0-py3-none-any.whl

Download URL jupyterlab_rtclauncher-0.4.0-py3-none-any.whl
Size 44.0 kB
Tags Python 3
SHA-256 checksum
How to use checksums
b1ea3bf500e449840e32c3123683739df0a89405dc68339c743e671e44e472bd
BLAKE2b-256 checksum
How to use checksums
20df06272184949e180a3b778d106d7be11968c2e64bca41b2f5a50d61e57d99
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.14.6

Security for jupyterlab-rtclauncher 0.4.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page