Running Notebooks inside a job

This page is about how to start a jupyter notebook on a GRICAD cluster node inside a OAR job.

Prerequisites:

Be sure to have previously configured your environment. In particular, you must :

  • have your SSH client configured to use proxy-command
  • have jupyter install in your current environment.

Configure jupyter, the init phase

The first time you use jupyter notebook on clusters, you have to configure it answering the questions from jupyter_config script. To do so, enter the following command : jupyter_config

Using notebooks

Once you have configured jupyter, you can launch or create a notebook on a compute node, entering the command : jupyter_start

Follow the script indications, and copy/paste the ssh tunnel command line on your host to enable SSH tunnel and access your jupyter notebook via your local browser (copy/paste the URL from the script).

Using tensorboard with tensorflow

If you need to use tensorboard to monitor your tensorflow jobs, you can do it with these few commands :

ActionCommand
start tensorboardtensorboard_start
check tensorboard statustensorboard_status
stop tensorboardtensorboard_stop

To access your tensorboard instance, copy/paste the URL from the tensorboard_start script in your web browser.

Note: jupyter_start script also opens the SSH tunnel to your tensorboard instance. No need to copy the URL from this script when you’re already working with jupyter and jupyter_start script.

If you have any question ro need any help on the subject, please mail to sos-gricad@univ-grenoble-alpes.fr.