workbench2.md

Install, upgrade workbench python packages

1. Terminal basics

Open Terminal

1. Click on the + Button to open Laucher
2. Click on the Terminal button on the Laucher plane
3. A new terminal tab will be open close to the Laucher tab
4. Click the terminal tab to interact with terminal

Show Quotas

In the terminal type:

kubectl describe quota

to see the information regarding your quota in Kubeflow.

2. Upgrade Jupyterlab Version

1. Click on Help menu button
2. Click on About JupyterLab to see the current JupyterLab version

The current juyterlab version is 3.4.3 from the default Kubeflow Jupyternotebook Image.

Lets upgrade it to version 3.6.7

In the opened terminal, you shall see the base conda env is activated. Lets type in:

# show python path
which python3;
# You shall see /opt/conda/bin/python3

# upgrade jupyterlab version
# we also need to upgrade the kfp package, due to some old dependency issues
python3 -m pip install --upgrade pip;
python3 -m pip install --user --upgrade jupyterlab==3.6.7 kfp==1.8.22;

Now close the Workbench/Jupyter Notebook Tab and restart the workbench server through stop and start

You can connect to the workbench again, by click on CONNECT

You can see the jupyterlab version by

1. Click on Help menu button
2. Click on About JupyterLab to see the current JupyterLab version

Now you shall see the JupyterLab is upgraded to 3.6.7

Important:

• it is important to notice, that the jupyterlab version is updated after the workbench server restart, since we used --user --upgrade flag during the upgrade process. More details about this please visit the next section.

3. Install and Upgrade Python packages

The docker file of Kubeflow image, you have choose sofar can be found at

• https://www.kubeflow.org/docs/components/notebooks/container-images/#kubeflow-images

3.1 Default Kubeflow Image Setup

You can find out the current conda env and pip package location by typing the following commands in terminal:

# show conda env
conda env list
# show conda pip package location using scipy as example
pip show scipy

The output is

Name: scipy
Version: 1.7.0
Summary: SciPy: Scientific Library for Python
Home-page: https://www.scipy.org
Author: 
Author-email: 
License: BSD
Location: /opt/conda/lib/python3.8/site-packages
Requires: numpy
Required-by: scikit-image, scikit-learn, seaborn, statsmodels

3.2 Install package not surviving restart

If we install or update a python packages using

# update pip
python3 -m pip install --upgrade pip;
# update the scipy package from 1.7.0 to 1.10.1
python3 -m pip install scipy==1.10.1
python3 -c 'import scipy; print(f"scipy.__version__: {scipy.__version__}")'

the new python packages will be installed in the folder /opt/conda/lib/python3.8/site-packages

Let's restart the current workbench and connect it again by

1. close the Workbench/Jupyter Notebook Tab and restart the workbench server through stop and start

then connect the Workbench and open terminal and typing:

python3 -c 'import scipy; print(f"scipy.__version__: {scipy.__version__}")';

it outputs

scipy.__version__: 1.7.0

Our previous upgraded python package scipy==1.10.1 is gone.

3.3 Install package surviving restart

To install or update python package surviving workbench restart, you need to use the --user --upgrade flag

python3 -m pip install --user --upgrade pip;
python3 -m pip install --user --upgrade scipy==1.10.1;
python3 -c 'import scipy; print(f"scipy.__version__: {scipy.__version__}")';

Let's restart the current workbench and connect it again by

1. close the Workbench/Jupyter Notebook Tab and restart the workbench server through stop and start

then connect the Workbench and open terminal and typing:

python3 -c 'import scipy; print(f"scipy.__version__: {scipy.__version__}")'

Now it show that the package is updated permanently

scipy.__version__: 1.10.1

By examining the location while typing the following command in the terminal:

pip show scipy

after scroll down the License output, which is very long, you will find:

...
Location: /home/jovyan/.local/lib/python3.8/site-packages

the location is now /home/jovyan/.local/lib/python3.8/site-packages instead of the prevous location /opt/conda/lib/python3.8/site-packages

Important:

• Install pip package with --user --upgrade flag to install it in the user partition /home/jovyan which is mounted as Workspace Volume to our workbench server
• Install pip package without --user flag, package will be installed in system partition of you workbench container, the system partition is not saved to image, after workbench pod restart the change in the system partition will be gone.

4. Stop and delete workbench

Important:

• If you don't stop the workbench, the allocated CPU resource will not be free.
• Let us stop the workbench/Notebook so that other users in Kubeflow can utilize these resources, this helps increase karma points :)

Type the following in your terminal in the current open workbench:

clear;
kubectl describe quota;

You shall see that we have CPU used:

Name:		kf-resource-quota
Namespace:		kubeflow-xxxx
Resource	Used	Hard
--------	----	----
basic-csi.storageclass.storage.k8s.io/persistentvolumeclaims	4	5
basic-csi.storageclass.storage.k8s.io/requests.storage	45Gi	50Gi
cpu	2300m	12

Your quota output may be different.

In this quota output example 2300m means 2.3 CPU time in seconds are used (Unit: cpu time in milliseconds). In this exampe, 2.3 CPUs on the kubernetes worker nodes at every seconds are being consumed.

4.1 Stop the workbench

1. close the Workbench Tab in your browser
2. stop the workbench by click on the stop button.

4.2 Delete the workbench

Since we don't need this jupyter notebook anymore, we can also delete it.

1. delete the workbench by click on the delete button after the workbench is stopped.

4.3 Delete the Workspace Volume

Navigate back to "kubeflow UI dashboard" -> "Volumes" sidebar menu

You can see that the created "test-volume" workspace is still there, even after you have deleted the workbench.

This "test-volume" can be used as workspace volume for new workbench to access the existing data inside this workspace volume.

Since it is not needed anymore in this tutorial, you can delete it by:

1. Click on delete button

Important:

• You noticed that in the "Used by" column of the test-volume there is no reference to a workbench, this indicate that the volume is not attached to any workbench or pod, it is safe to delete.
• You have a quota of "5" volume in storage class homedir by default in the namespace, remove the volume you don't need, or contact your kubeflow admin to get more quota for the volume.