dariuszparys / azureml-terraform-template

This repo hosts the code to roll out entire infrastructure for Azure ML with or without private endpoints

Azure Machine Learning Deployment

This template deploys complete Azure Machine Learning enviornment via Terraform with very minimal inputs. It supports two kinds of following Azure Machine Learning deployments.

• Azure ML with private endpoints and VNet binding
• Azure ML without private endpoints and VNet binding.

With a single switch of terraform parameter you can choose which kind of AML architecture you want to deploy. The details of these deployments are discussed in the next sections.

Architecture for Azure ML Resources With Private Endpoints and VNet Binding

This includes deployment of the following resources:

• Azure Machine Learning Workspace with Private Link.
• Azure Storage Account with Private Link for Blob and File and VNET binding with Service Endpoints.
• Azure Key Vault with Private Link for Vault and VNET binding with Service Endpoints.
• Azure Container Registry (Private endpoints and VNET binding is not enabled as this feature needs to be requested from MSFT).
• Azure Application Insights.
• DSVM (Windows) for development and testing (in VNET and DSVMSubnet).
• DSVM is setup to auto shutdown at 18:00 PST to save on costs.
• Azure Bastion Host as an entry point to access the locked down enviornment via DSVM.
• One Virtual Network.
• AMLSubnet for AML Compute and AML resources such as Storage and KeyVault.
• DSVMSubnet for DSVM.
• AzureBastionSubnet for Bastion Host.
• Network Security group for DSVMSubnet which accepts incoming traffic from Bastion Host only.
• Network Security group for AzureVMSubnet which accepts incoming traffic from AzureCloud, GatewayManager and HTTPs.
• Managed AML Compute Cluster with scale up to 4 VMs (in VNET and AMLSubnet).
• Creates timestamp, create-by and environment Tags to all resources.

Architecture for Azure ML Resources Without Private Endpoints and VNet Binding

This includes deployment of the following resources:

• Azure Machine Learning Workspace.
• Azure Storage Account.
• Azure Key Vault.
• Azure Container Registry.
• Azure Application Insights.
• DSVM (Windows) for development and testing (in VNET and DSVMSubnet).
• DSVM is setup to auto shutdown at 18:00 PST to save on costs.
• Azure Bastion Host as an entry point to access the locked down enviornment via DSVM.
• One Virtual Network.
• AMLSubnet for AML Compute Cluster.
• DSVMSubnet for DSVM.
• AzureBastionSubnet for Bastion Host.
• Network Security group for DSVMSubnet which accepts incoming traffic from Bastion Host only.
• Network Security group for AzureVMSubnet which accepts incoming traffic from AzureCloud, GatewayManager and HTTPs.
• Managed AML Compute Cluster with scale up to 4 VMs (in VNET and AMLSubnet).
• Creates timestamp, create-by and environment Tags to all resources.

Prerequisites

• Install Azure CLI
• Install Azure CLI ML extension by running az extension add -n azure-cli-ml
• Install Terraform

Run Deployment

Navigate to deployment\terraform\aml

• Run terraform init
• Run terraform plan
• Run terraform apply

You will need to provide values for project_code and env_code. Value for location is set to West US 2 and enable_private_link is set to true.