Jupyter on CGD Systems

How to run JupyterLab on a CGD server and access it from a local browser via an SSH tunnel.

Prerequisites

  • A working conda (or mamba) environment on the CGD system with jupyterlab installed.
  • VPN access if connecting from outside the UCAR network.

Overview

  • If remote: connect to VPN
  • SSH to a CGD server (magnolia, topaz, andre, thorodin, etc.)
  • Activate your conda environment and navigate to your working directory
  • Start the Jupyter server
  • Open an SSH tunnel from your local machine to the server
  • Open JupyterLab in your local browser

Step-by-Step

Terminal 1 — on the CGD server 
ssh [email protected]
conda activate your_environment
cd /your/code/repository
jupyter lab --port=8899 --no-browser
Port note: The default Jupyter port is 8888, but it is often already in use. Port 8899 is a convenient alternative — just pick any unused port and use it consistently in both commands.
Terminal 2 — on your local machine 
ssh -N -f -L 8899:localhost:8899 [email protected]
Browser

Navigate to:

localhost:8899

You will likely see an authentication page asking for a token:

Jupyter login page

Go back to Terminal 1 (where the Jupyter server is running) to find the token. The output will look something like this:

Terminal showing running Jupyter server

Copy the token and paste it into the browser field. JupyterLab will open. Leave both terminal sessions running while you work.

Shutting Down

In JupyterLab, use File > Shut Down to stop the server, or press Ctrl+C in Terminal 1 and confirm when prompted.

To close the SSH tunnel, run the following in any local terminal:

ps aux | grep 8899
kill -9 [PID]