Jupyter Lab

Note

This guide is applicable only to those who requested Jupyter Lab. The examples use roselab1, but the same principles apply to roselab2~4.

Quick Start

Setting Up Permanent Port Mapping

To set up a permanent HTTPS port mapping for JupyterLab using the common utilities:

1. Navigate to the common utilities directory:

   cd /public/common-utilities/

2. Run the client script:

   python client.py

3. Choose option 1: "Add or delete port mapping"

4. Select "Add a new port mapping"

5. When prompted, enter the following details:

   • Source port: Choose an available port within your assigned port range (e.g., 25308, if your range is 25300-25399)
   • Destination port: 8888 (JupyterLab's default port)
   • Protocol: HTTPS

6. Confirm your choices when prompted

After setting up the HTTPS mapping, you can access JupyterLab securely via https://roselab1.ucsd.edu:25308 (replace 25308 with your chosen port number).

Note

HTTPS mapping adds a security layer and is more browser-friendly, but it only supports hosted HTTP web services.

SSH Port Forwarding

If you didn't request port mapping, you'll need to use SSH port forwarding to access your Jupyter service. After confirming successful SSH access to the server:

1. If you've configured your .ssh/config file:

   laptop$ ssh -fN -L 8888:localhost:8888 roselab1

2. If you haven't configured .ssh/config:

   laptop$ ssh -fN -L 8888:localhost:8888 ubuntu@roselab1.ucsd.edu -p <ssh-port> -i path/to/keyfile

These commands initiate a background connection forwarding your local port 8888 to the remote Jupyter service. Access your server at http://localhost:8888. You may need to rerun the command if the network environment changes.

Using Visual Studio Code

If you're using Visual Studio Code with the Remote SSH extension:

1. Connect to your RoseLab server.
2. Go to the "Ports" view in the left or bottom sidebar.
3. Click on "Forward a Port" and enter 8888.
4. Access Jupyter Lab at http://localhost:8888.

Troubleshooting Port Occupancy

If you encounter a "port already in use" error:

For macOS/Linux:

sudo lsof -i :8888

For Windows (PowerShell):

Get-NetTCPConnection -LocalPort 8888

These commands will show processes using port 8888. You can then close the conflicting process or choose a different port.

Configuration

Jupyter Lab runs as a service on localhost:8888, configured in /etc/systemd/system/jupyterlab.service. If killed, it will restart within seconds.

To change the default password, edit ~/.jupyter/jupyter_lab_config.py:

c.ServerApp.token = 'your-new-password'

Save the file and run sudo systemctl restart jupyterlab to apply changes.

WARNING

Use a strong password, especially for publicly accessible services. Jupyter Lab will remember your login.

Adding Environments

Jupyter Lab uses the Anaconda base environment. To add a new environment:

(base)$ conda create -n py38 python=3.8
(base)$ conda activate py38
(py38)$ conda install -c anaconda ipykernel
(py38)$ python -m ipykernel install --user --name=py38

To remove an environment's kernel configuration:

rm -rf /home/ubuntu/.local/share/jupyter/kernels/py38

Debugging

If Jupyter Lab malfunctions, check the service log:

$ journalctl -u jupyterlab -n 100

Common issues include out-of-memory errors. Monitor your GPU and memory usage regularly at Grafana.

Note

Replace roselab1 with your assigned server (roselab2, roselab3, or roselab4) in all examples.