Deploying Charmed MLflow v2 and Kubeflow to EKS

charm
, ,
1

In this guide you will create an AWS EKS cluster, setup Juju on it, deploy the MLflow and Kubeflow bundles and relate them to each other.

Requirements:

Deploy EKS cluster

See here for a complete guide on how to do exactly that. Do not forget to edit the instanceType field under managedNodeGroups[0].instanceType from t2.2xlarge to t3.2xlarge, as instructed in the guide, since worker nodes of type t3.2xlarge are required for deploying both MLFlow and Kubeflow.

Setup Juju

First, install Juju with:

sudo snap install juju --classic

Next, connect it to the Kubernetes (we chose the name kubeflow, but you can choose any other name).

juju add-k8s kubeflow

Now create the controller (we chose the name kubeflow-controller, but you can choose any other name).

juju bootstrap --no-gui kubeflow kubeflow-controller

Create a Juju model by running:

juju add-model kubeflow

Deploy Mlflow bundle

Now deploy MLflow bundle with the following command:

juju deploy mlflow --channel=2.15/stable --trust

Wait until all charms are in active state. You can check the state of the charms with:

juju status --watch 5s

Deploy Kubeflow

Now deploy Kubeflow bundle with the following command:

juju deploy kubeflow --channel=1.9/stable --trust

Wait until all charms are in active state. You can check the state of the charms with:

juju status --watch 5s

Relate MLflow to Kubeflow

To run MLflow seamlessly with Kubeflow, deploy the resource-dispatcher charm with:

juju deploy resource-dispatcher --channel edge --trust

Resource dispatcher is responsible for configuring MLflow to relate Kubernetes objects for Kubeflow user namespaces. Relate resource dispatcher to the MLflow server with the following commands:

juju relate mlflow-server:secrets resource-dispatcher:secrets
juju relate mlflow-server:pod-defaults resource-dispatcher:pod-defaults

Wait until all charms are in active state. You can check the state of the charms with command:

juju status --watch 5s

Configure Kubeflow dashboard

To enable authentication for the dashboard, set a username and password as follows:

juju config dex-auth static-password=user123
juju config dex-auth static-username=user123@email.com

To find the IP address associated with your deployment, run the following command:

microk8s kubectl -n kubeflow get svc istio-ingressgateway-workload -o jsonpath='{.status.loadBalancer.ingress[0].ip}'

You should see an output like this: 10.64.140.43, which is the default IP address used in the MicroK8s configuration.

To access your deployment, open a browser and visit the dashboard IP. You should see the login page where you need to enter the credentials previously set up.

2

Archiving this article, since it has been replaced by this guide.

3