Simple Helm Chart for ML model deployments

Introduction

This Helm chart installs a serialized ML model API for predictions in a Kubernetes cluster.

Prerequisites

• Kubernetes cluster 1.22+
• Helm 3.12.0+

Installation

Add Helm repository

helm repo add ml-service https://github.com/degrasse-python/ml-service
helm repo update

Configure the chart

The following items can be set via --set flag during installation or configured by editing the values.yaml directly (need to download the chart first).

Configure the way how to expose mlflow service:

• Ingress: The ingress controller must be installed in the Kubernetes cluster.
• ClusterIP: Exposes the service on a cluster-internal IP. Choosing this value makes the service only reachable from within the cluster.
• NodePort: Exposes the service on each Node’s IP at a static port (the NodePort). You’ll be able to contact the NodePort service, from outside the cluster, by requesting NodeIP:NodePort.
• LoadBalancer: Exposes the service externally using a cloud provider’s load balancer.

Install the chart

Install the ml-service helm chart with a release name my-release:

helm install --name my-release ml-service/ml-service

Uninstallation

To uninstall/delete the my-release deployment:

helm delete --purge my-release

Configuration

The following table lists the configurable parameters of the mlflow chart and the default values.

Parameter	Description	Default
Image
image.repository	MLFlow Image name	ayadi05/mlflow
image.tag	MLFlow Image tag	1.5.0
image.pullPolicy	MLFlow Image pull policy	IfNotPresent
Service
service.type	Type of service for MLFlow frontend	ClusterIP
service.port	Port to expose service	443
service.targetPort	Port where the service is reachable	8080
service.tls	tls if service type is enabled	true
Ingress
ingress.enabled	Enables Ingress	false
ingress.annotations	Ingress annotations	{}
ingress.path	Path to access mlapi	/
ingress.hosts	Ingress hosts	[]
ingress.tls	Ingress TLS configuration	[]
Resources
resources	CPU/Memory resource requests/limits	{}

Contributing

At the moment we are not taking any pull requests because this is just an example for folks to use for inspo.

License

MIT