Integrate Teradata Jupyter extensions with Google Vertex AI

This how-to shows you how to add Teradata Extensions to a Jupyter Notebooks environment. A hosted version of Jupyter Notebooks integrated with Teradata Extensions and analytics tools is available for functional testing for free at https://clearscape.teradata.com.

Overview

Teradata Jupyter extensions provide Teradata SQL kernel and several UI extensions to allow users to easily access and navigate Teradata database from Jupyter envioronment. Google Vertex AI is Google Cloud’s new unified ML platform. Vertex AI Workbench provides a Jupyter-base development environment for the entire data science workflow. This article describes how to integate our Jupyter extensions with Vertex AI Workbench so that Vertex AI users can take advantage of our Teradata extensions in their ML pipeline.

Vertex AI workbench supports two types of notebooks: managed notebooks and user-managed notebooks. Here we will focus on user-managed notebooks. We will show two ways to integrate our Jupyter extensions with user-managed notebooks: use startup script to install our kernel and extensions or use custom container.

Prerequisites

  • Access to a Teradata Vantage instance

    If you need a test instance of Vantage, you can provision one for free at https://clearscape.teradata.com.
  • Google Cloud account with Vertex AI enabled

  • Google cloud storage to store startup scripts and Teradata Jupyter extension package

Integration

There are two ways to run Teradata Jupyter Extensions in Vertex AI:

These two integration methods are described below.

Use startup script

When we create a new notebook instance, we can specify a startup script. This script runs only once after the instance is created. Here are the steps:

  1. Download Teradata Jupyter extensions package

    Go to Vantage Modules for Jupyter page to download the Teradata Jupyter extensions package bundle Linux version.

  2. Upload the package to a Google Cloud storage bucket

  3. Write a startup script and upload it to cloud storage bucket

    Below is a sample script. It fetches Teradata Jupyter extension package from cloud storage bucket and installs Teradata SQL kernel and extensions.

    #! /bin/bash
    
    cd /home/jupyter
    mkdir teradata
    cd teradata
    gsutil cp gs://teradata-jupyter/* .
    unzip teradatasql*.zip
    
    # Install Teradata kernel
    cp teradatakernel /usr/local/bin
    
    jupyter kernelspec install ./teradatasql --prefix=/opt/conda
    
    # Install Teradata extensions
    pip install --find-links . teradata_preferences_prebuilt
    pip install --find-links . teradata_connection_manager_prebuilt
    pip install --find-links . teradata_sqlhighlighter_prebuilt
    pip install --find-links . teradata_resultset_renderer_prebuilt
    pip install --find-links . teradata_database_explorer_prebuilt
    
    # PIP install the Teradata Python library
    pip install teradataml
    
    # Install Teradata R library (optional, uncomment this line only if you use an environment that supports R)
    #Rscript -e "install.packages('tdplyr',repos=c('https://r-repo.teradata.com','https://cloud.r-project.org'))"
  4. Create a new notebook and add the startup script from cloud storage bucket

    create a new notebook with startup script

  5. It may take a few minutes for the notebook creation process to complete. When it is done, click on Open notebook.

    Open notebook

Use custom container

Another option is to provide a custom container when creating a notebook.

  1. Download Teradata Jupyter extensions package

    Go to Vantage Modules for Jupyter page to download the Teradata Jupyter extensions package bundle Linux version.

  2. Copy this package to your work directory and unzip it

  3. Build custom Docker image

    The custom container must expose a service on port 8080. It is recommended to create a container derived from a Google Deep Learning Containers image, because those images are already configured to be compatible with user-managed notebooks.

    Below is a sample Dockerfile you can use to build a Docker image with Teradata SQL kernel and extensions installed:

    # Use one of the deep learning images as base image
    # if you need both Python and R, use one of the R images
    FROM gcr.io/deeplearning-platform-release/r-cpu:latest
    
    USER root
    
    ##############################################################
    # Install kernel and copy supporting files
    ##############################################################
    
    # Copy the kernel
    COPY ./teradatakernel /usr/local/bin
    
    RUN chmod 755 /usr/local/bin/teradatakernel
    
    # Copy directory with kernel.json file into image
    COPY ./teradatasql teradatasql/
    
    # Copy notebooks and licenses
    COPY ./notebooks/ /home/jupyter
    COPY ./license.txt /home/jupyter
    COPY ./ThirdPartyLicenses/ /home/jupyter
    
    # Install the kernel file to /opt/conda jupyter lab instance
    RUN jupyter kernelspec install ./teradatasql --prefix=/opt/conda
    
    ##############################################################
    # Install Teradata extensions
    ##############################################################
    
    RUN pip install --find-links . teradata_preferences_prebuilt && \
        pip install --find-links . teradata_connection_manager_prebuilt && \
        pip install --find-links . teradata_sqlhighlighter_prebuilt && \
        pip install --find-links . teradata_resultset_renderer_prebuilt && \
        pip install --find-links . teradata_database_explorer_prebuilt
    
    # Give back ownership of /opt/conda to jovyan
    RUN chown -R jupyter:users /opt/conda
    
    # PIP install the Teradata Python libraries
    RUN pip install teradataml
    
    # Install Teradata R library (optional, include it only if you use a base image that supports R)
    RUN Rscript -e "install.packages('tdplyr',repos=c('https://r-repo.teradata.com','https://cloud.r-project.org'))"
  4. In your work directory (where you unzipped Teradata Jupyter extensions package), run docker build to build the image:

    docker build -f Dockerfile imagename:imagetag .
  5. Push the docker image to Google container registry or artifact registry

    Please refer to the following documentations to push docker image to registry:

  6. Create a new notebook

    In Environment section, set custom container field to the location of your newly created custom container:

    Open notebook

Did this page help?