Provisioning an MLOps Platform in the Cloud

Now that you have an understanding of MLOps and the different stages of a machine learning (ML) life cycle, in this chapter, you will provision a managed Red Hat OpenShift cluster on the Amazon Web Services (AWS) cloud. You will then provision Red Hat OpenShift Data Science (ODS) and partner components on the Red Hat OpenShift platform.

The focus of this chapter is to provide you with an overview of how to build your MLOps platform using OpenShift and a cloud vendor. The agility of Red Hat OpenShift paired with cloud services provides a solid foundation for you to build your MLOps platform in very little time. Keep in mind that the OpenShift platform is cloud-agnostic, and you can use it with your on-premises infrastructure if this is the path you want to take.

This chapter does not make you an expert in provisioning the OpenShift platform. There are many books and tons of documentation providing such details, and we leave it to you...

Technical requirements

You will need a computer with internet access to provision the platform in AWS. You will also need an active AWS account, which you will use to create the resources required by the platform. You should have basic knowledge of interacting with the AWS portal.

Installing OpenShift on AWS

Red Hat OpenShift Service on AWS (ROSA) is a fully managed Red Hat OpenShift platform. ROSA is operated and supported jointly by AWS and Red Hat. With ROSA, you move your focus from managing infrastructure to managing applications and bringing more business value to your organization. ROSA provides an integrated experience for OpenShift cluster creation, a pay-as-you-go (hourly and annual) billing, and a single invoice for AWS deployments. ROSA helps to reduce operational complexity with automated deployment and management, backed by a global Red Hat site reliability engineering (SRE) team.

The workflow for ROSA creation involves many steps. The setup defined here shows you the steps involved in an easy way and for learning purposes. The settings provided here are not recommended for production clusters.

Note

In a production setup, you will need to integrate the platform with your own identity provider (IdP). You may need to configure firewalls...

Installing Red Hat ODS

The Red Hat ODS service enables data science teams to execute data science and ML workflows by integrating Red Hat components, open source software (OSS), and partner offerings. Its primary goal is to provide a collaborative and scalable environment for data scientists and data engineers to develop, deploy, and manage ML and AI applications.

With ODS, data scientists can leverage familiar tools such as Jupyter Notebook to create interactive development environments for their data analysis and model development tasks. You can easily build and train ML models using frameworks such as TensorFlow, PyTorch, and scikit-learn. Red Hat ODS includes JupyterHub, Git integration, model deployment, and model serving and is based on the upstream open source project, Open Data Hub (ODH). You will learn more about the details and internal components of Red Hat ODS in the next chapters.

OpenShift makes it very easy to install ODS in the following easy steps. ODS is packaged...

Installing partner software on RedHat ODS

In order to complete our MLOps platform, we will need to install additional tools to OpenShift to complement the features of ODS. Several tools are considered partner software of the ODS platform. These software products are listed in the ODS console and can be viewed by clicking the Explore menu item in the ODS console, as shown in Figure 2.34:

Figure 2.34 – ODS console showing partner software

One of the things that you will need to complete the MLOps platform is data versioning, and you will need data lineage too. We will use Pachyderm for this need.

Pachyderm is a powerful OSS tool designed to manage data in modern data pipelines. It serves as a data versioning and lineage system, enabling efficient tracking and control of data changes throughout the data processing life cycle.

With Pachyderm, you can easily keep track of modifications made to your data, similar to how version control systems ...

Installing Pachyderm

Because Pachyderm is a Kubernetes-native platform, it can also natively run on OpenShift. Pachyderm is available as a Kubernetes operator in OperatorHub.

The following steps will guide you through the operator installation of Pachyderm:

1. In the OpenShift console, navigate to Operators -> OperatorHub.
2. Search for Pachyderm. You should see Pachyderm shown as a tile, as shown in Figure 2.35:

Figure 2.35 – OperatorHub showing Pachyderm

3. If you are presented with two options, as shown in Figure 2.35, choose the one tagged as Certified and then click the Install button. Certified operators are operators that are verified by Red Hat to run in OpenShift.
4. Use the default options in the Install Operator screen, as shown in Figure 2.36, and then click Install:

Figure 2.36 – Install Operator dialog of Pachyderm

5. Once the installation is complete, Pachyderm should be...

Summary

In this chapter, you have learned how to provision the platform. You have installed the Red Hat OpenShift platform, Red Hat ODS, and Pachyderm. You have seen how the MLOps components defined in the previous chapters are translated into the software components presented in this chapter. It gives you a mental model of what a complete MLOps platform would look like.

In the next few chapters, you will dive deeper into each of these components and understand how they enable your team to move forward. Now that you have learned how to install Kubernetes operators from OperatorHub. We will also be adding more open source components to your MLOps technology stack as you go through the later chapters.