BentoML

Build, Ship, and Scale AI Applications

BentoML is an open-source platform for building, shipping, and scaling AI applications. Starting with the release of Kubeflow 1.7, BentoML provides a native integration with Kubeflow through Yatai. This integration allows you to package models trained in Kubeflow notebooks or pipelines, and deploy them as microservices in a Kubernetes cluster through BentoML’s cloud native components and custom resource definitions (CRDs). This documentation provides a comprehensive guide on how to use BentoML and Kubeflow together to streamline the process of deploying models at scale.

In this tutorial, we will deploy a fraud detection application that determines whether a transaction is fraudulent by simultaneously calling three models trained with the Kaggle IEEE-CIS Fraud Detection dataset in the Kubeflow notebook. We have already built and uploaded the application as a Bento, an archive with all the source code, models, data files and dependency configurations required for running the application. If you’re interested in learning how to train the models and build the Bento, see the Fraud Detection on Kubeflow example project. Here we will focus on the two deployment workflows using BentoML’s Kubernetes operators: deploying directly from the Bento, and deploying from an OCI image built from the Bento.

All source code for model training, service definition, and Kubernetes deployment can be found in the Fraud Detection on Kubeflow example.

image

Prerequisites

Before starting this tutorial, make sure you have the following:

Deploy to Kubernetes Cluster

BentoML offers three custom resource definitions (CRDs) in the Kubernetes cluster.

  • BentoRequest - Describes the metadata needed for building the container image of the Bento, such as the download URL. Created by the user and reconciled by the yatai-image-builder operator.
  • Bento - Describes the metadata for the Bento such as the address of the image. Created by users or by the yatai-image-builder operator after reconciling the BentoRequest resources.
  • BentoDeployment - Describes the metadata of the deployment such as resources and autoscaling behaviors. Reconciled by the yatai-deployment operator to create Kubernetes deployments of API Servers and Runners.

Next, we will demonstrate two ways of deployment.

  1. Deploying using a BentoRequest resource by providing a Bento
  2. Deploying Using a Bento resource by providing a pre-built container image from a Bento

Deploy with BentoRequest CRD

In this workflow, we will use the yatai-image-builder operator to download and containerize the Bento and push the resulting OCI image to a remote registry. We will then use the yatai-deployment operator to deploy the containerized Bento image.

Apply the BentoRequest and BentoDeployment resources as defined in deployment_from_bentorequest.yaml.

curl -s https://raw.githubusercontent.com/bentoml/BentoML/main/examples/kubeflow/deployment_from_bentorequest.yaml | kubectl apply -f -

Once the resources are created, the yatai-image-builder operator will reconcile the BentoRequest resource and spawn a pod to download and build the container image from the provided Bento defined in the resource. The yatai-image-builder operator will push the built image to the container registry specified during the installation and create a Bento resource with the same name. At the same time, the yatai-deployment operator will reconcile the BentoDeployment resource with the provided name and create Kubernetes deployments of API Servers and Runners from the container image specified in the Bento resource.

Deploy with Bento CRD

In this workflow, we will build and push the container image from the Bento. We will then leverage the yatai-deployment operator to deploy the containerized Bento image.

Download and import the Bento for the Fraud Detection application, fraud_detection.bento.

curl -k https://bentoml.com.s3.amazonaws.com/kubeflow/fraud_detection.bento -o fraud_detection.bento
bentoml import fraud_detection.bento

Containerize the image through containerize sub-command.

bentoml containerize fraud_detection:o5smnagbncigycvj -t your-username/fraud_detection:o5smnagbncigycvj

Push the containerized Bento image to a remote repository of your choice.

docker push your-username/fraud_detection:o5smnagbncigycvj

Update the image to the address of your repository and apply the Bento and BentoDeployment resources as defined in deployment_from_bento.yaml.

curl -s https://raw.githubusercontent.com/bentoml/BentoML/main/examples/kubeflow/deployment_from_bento.yaml | kubectl apply -f -

Once the resources are created, the yatai-deployment operator will reconcile the BentoDeployment resource with the provided name and create Kubernetes deployments of API Servers and Runners from the container image specified in the Bento resource.

Verify Deployment

Verify the deployment of API Servers and Runners. Note that API server and runners are run in separate pods and created in separate deployments that can be scaled independently.

kubectl -n kubeflow get pods -l yatai.ai/bento-deployment=fraud-detection

NAME                                        READY   STATUS    RESTARTS   AGE
fraud-detection-67f84686c4-9zzdz            4/4     Running   0          10s
fraud-detection-runner-0-86dc8b5c57-q4c9f   3/3     Running   0          10s
fraud-detection-runner-1-846bdfcf56-c5g6m   3/3     Running   0          10s
fraud-detection-runner-2-6d48794b7-xws4j    3/3     Running   0          10s

Port forward the Fraud Detection service to test locally. You should be able to visit the Swagger page of the service by requesting http://0.0.0.0:8080 while port forwarding.

kubectl -n kubeflow port-forward svc/fraud-detection 8080:3000 --address 0.0.0.0

Conclusion

Congratulations! You completed the example. Let’s recap what we have learned.

  • Trained three fraud detection models and saved them to the BentoML model store.
  • Created a BentoML service that runs inferences on all three models simultaneously, combines and returns the results.
  • Containerized the BentoML service into an OCI image and pushed the image to a remote repository.
  • Created BentoML CRDs on Kubernetes to deploy the Bento in a microservice architecture.

Additional Resources

Feedback

Was this page helpful?


Last modified March 22, 2023: Update the BentoML documentation (efbf4b1)