Build and deploy data and machine learning pipelines with Flyte on Azure Kubernetes Service (AKS)

Article
08/01/2024

This article shows you how to use Flyte on Azure Kubernetes Service (AKS). Flyte is an open-source workflow orchestrator that unifies machine learning, data engineering, and data analytics stacks to help you build robust and reliable applications. When using Flyte as a Kubernetes-native workflow automation tool, you can focus on experimentation and providing business value without increasing your scope to infrastructure and resource management. Keep in mind that Flyte isn't officially supported by Microsoft, so use it at your own discretion.

For more information, see Introduction to Flyte.

Important

Open-source software is mentioned throughout AKS documentation and samples. Software you deploy is excluded from AKS service-level agreements, limited warranty, and Azure support. As you utilize open-source technology alongside AKS, consult the support options available from the respective communities and project maintainers to develop a plan.

For example, Ray's GitHub repository describes several platforms that vary in response time, purpose, and support level.

Microsoft takes responsibility for building the open-source packages we deploy on AKS. That responsibility includes having complete ownership of the build, scan, sign, validate, and hotfix process, as well as control over the binaries in container images. For more information, see Vulnerability management for AKS and AKS support coverage.

Flyte use cases

Flyte can be used for a variety of use cases, including:

Deliver models for streamlined profit and loss financial calculations.
Process petabytes of data to efficiently conduct 3D mapping of new areas.
Quickly rollback to previous versions and minimize impact of bugs in your pipelines.

For more information, see Core Flyte use cases.

Prerequisites

An Azure subscription. If you don't have an Azure subscription, you can create a free account.
- If you have multiple subscriptions, make sure you select the correct one using the az account set --subscription <subscription-id> command.
The Azure CLI installed and configured. Check your version using the az --version command. If you need to install or upgrade, see Install the Azure CLI.
The Helm CLI installed and updated. Check your version using the helm version command. If you need to install or upgrade, see Install Helm.
The kubectl CLI installed and updated. Install it locally using the az aks install-cli command or using Install kubectl.
A local Docker development environment. For more information, see Get Docker.
flytekit and flytectl installed. For more information, see Flyte installation.

Note

If you're using the Azure Cloud Shell, the Azure CLI, Helm, and kubectl are already installed.

Set environment variables

Set environment variables for use throughout the article. Replace the placeholder values with your own values.

export RESOURCE_GROUP="<resource-group-name>"
export LOCATION="<location>"
export CLUSTER_NAME="<cluster-name>"
export DNS_NAME_PREFIX="<dns-name-prefix>"

Create an AKS cluster

Create an Azure resource group for the AKS cluster using the az group create command.
```
az group create --name $RESOURCE_GROUP --location $LOCATION
```

Create an AKS cluster using the az aks create command with the --enable-azure-rbac, --enable-managed-identity, --enable-aad, and --dns-name-prefix parameters.

az aks create --resource-group $RESOURCE_GROUP --name $CLUSTER_NAME --enable-azure-rbac --enable-managed-identity --enable-aad --dns-name-prefix $DNS_NAME_PREFIX  --generate-ssh-keys

Connect to your AKS cluster

Configure kubectl to connect to your AKS cluster using the az aks get-credentials command.
```
az aks get-credentials --resource-group $RESOURCE_GROUP --name $CLUSTER_NAME
```

Add the Flyte Helm repository

Add the Flyte Helm repository using the helm repo add command.
```
helm repo add flyteorg https://flyteorg.github.io/flyte
```

Find Flyte Helm charts

Search for Flyte Helm charts using the helm search repo command.

helm search repo flyteorg

The following example output shows some of the available Flyte Helm charts:

NAME                    CHART VERSION   APP VERSION     DESCRIPTION
flyteorg/flyte          v1.12.0                         A Helm chart for Flyte Sandbox
flyteorg/flyte-binary   v1.12.0         1.16.0          Chart for basic single Flyte executable deployment
flyteorg/flyte-core     v1.12.0                         A Helm chart for Flyte core
flyteorg/flyte-deps     v1.12.0                         A Helm chart for Flyte dependencies
flyteorg/flyte-sandbox  0.1.0           1.16.1          A Helm chart for the Flyte local sandbox
flyteorg/flyteagent     v0.1.10                         A Helm chart for Flyte Agent

Update the repository using the helm repo update command.
```
helm repo update
```

Deploy a Flyte chart on AKS

In this section, you deploy the flyte-binary Helm chart so you can begin building and deploying data and machine learning pipelines with Flyte on AKS. The flyte-binary chart is a basic single Flyte executable deployment.

Create a namespace for your Flyte deployment using the kubectl create namespace command.
```
kubectl create namespace <namespace-name>
```
Install a Flyte Helm chart using the helm install command. In this example, we use the flyte-binary chart.
```
helm install flyte-binary flyteorg/flyte-core --namespace <namespace-name>
```

Verify that the Flyte deployment is running using the kubectl get services command.

kubectl get services --namespace <namespace-name> --output wide

The following condensed example output shows the Flyte deployment:

NAME                            TYPE           CLUSTER-IP     EXTERNAL-IP     PORT(S)          AGE
flyteorg-flyte-binary-grpc      ClusterIP      xx.x.xx.xxx    <none>          81/TCP           1m
flyteorg-flyte-binary-http      ClusterIP      xx.x.xx.xxx    <none>          80/TCP           1m
flyteorg-flyte-binary-webhook   ClusterIP      xx.x.xx.xxx    <none>          80/TCP           1m

Next steps

In this article, you learned how to install Flyte on AKS using a Helm chart. The Flyte project also maintains a reference implementation for AKS that automatically configures all the dependencies and deploys a production grade Flyte cluster.

To start building and deploying data and machine learning pipelines, see the following articles:

Share via