Here’s a Python script that will:
-
Start a JupyterLab instance in the background on your cluster.
-
Log the token, start time, job id (i.e., the PID of the Jupyter process), and port to a file.
-
Allow you to later copy the log to your local machine and kill the JupyterLab process using the PID (job id).
Assumptions: You're running this script on a Linux cluster machine.
-
Python and JupyterLab are installed.
-
No job scheduler is used.
-
You're using SSH port forwarding to connect from your local machine (I'll note how below).
In the cluster, run the python script using nohup and &.
nohup python3 start_jupyter_lab.py > ~/jupyter_wrapper.log 2>&1 &
Run the python file after copying it to the cluster account location. Change the location of the log file accordingly.
In the local system:
ssh -N -L 8888:localhost:<assigned_port> your_user@cluster_hostname
Find the port details from the json file and replace it at <assigned_port>.
Now in your prefered browser, start using the jupyter server by typing the address,
http://localhost:8888/?token=abc123...
Hope this helps.
Author: Soumyaranjan Dash