# Title, CIDR ranges, arrow labels and flow steps for the gallery diagram. Use with: # terravision draw --source azure-ml-mlops.tvg.json --annotate azure-ml-mlops.annotations.yml format: '0.3' title: Azure Machine Learning MLOps with private workspace, CI runners and AKS inference update: azurerm_virtual_network.ml: address_space: - 10.64.0.0/16 azurerm_subnet.gateway: address_prefixes: - 10.64.0.0/27 label: GatewaySubnet azurerm_subnet.ci_runners: address_prefixes: - 10.64.1.0/26 azurerm_subnet.training: address_prefixes: - 10.64.4.0/22 azurerm_subnet.inference: address_prefixes: - 10.64.8.0/22 azurerm_subnet.app_integration: address_prefixes: - 10.64.12.0/26 azurerm_subnet.private_endpoints: address_prefixes: - 10.64.20.0/24 azurerm_linux_virtual_machine_scale_set.github_runners: label: GitHub Runners azurerm_machine_learning_compute_cluster.training: label: GPU Training Cluster azurerm_machine_learning_compute_instance.notebooks: label: Compute Instances azurerm_machine_learning_workspace.ml: label: Azure ML Workspace azurerm_linux_web_app.claims: label: Claims App azurerm_storage_container.training_data: label: Training Data azurerm_storage_container.models: label: Model Artifacts tv_azurerm_internet.github: label: GitHub azurerm_container_registry.images: label: Container Registry azurerm_virtual_network_gateway.vpn: label: VPN Gateway azurerm_kubernetes_cluster.inference: label: AKS Inference azurerm_application_insights.ml: label: Application Insights azurerm_log_analytics_workspace.ml: label: Log Analytics azurerm_private_dns_zone.privatelink: label: Private DNS Zones connect: tv_azurerm_users.data_scientists: - azurerm_virtual_network_gateway.vpn: Point-to-site VPN azurerm_public_ip.egress: - tv_azurerm_internet.github: HTTPS 443 azurerm_linux_web_app.claims: - azurerm_kubernetes_cluster.inference: Score HTTPS flows: Train: description: A merged change retrains and registers the model steps: - resource: tv_azurerm_users.data_scientists -> azurerm_virtual_network_gateway.vpn xlabel: Connect detail: Data scientists connect over point-to-site VPN - resource: azurerm_virtual_network_gateway.vpn -> azurerm_machine_learning_compute_instance.notebooks xlabel: Explore detail: and explore data on their compute instances - resource: azurerm_public_ip.egress -> tv_azurerm_internet.github xlabel: Pick up detail: A merge to main queues a workflow; the self-hosted runners pick it up from GitHub through the NAT gateway - resource: azurerm_linux_virtual_machine_scale_set.github_runners -> azurerm_private_endpoint.workspace xlabel: Submit detail: The runners submit the training pipeline to the Azure ML workspace through its private endpoint - resource: azurerm_machine_learning_workspace.ml -> azurerm_machine_learning_compute_cluster.training xlabel: Train detail: The workspace scales out the GPU compute cluster for the run - resource: azurerm_machine_learning_compute_cluster.training -> azurerm_private_endpoint.blob xlabel: Data detail: which reads the training data and writes the evaluated model to Blob Storage, then registers it Deploy and score: description: The approved model serves claims steps: - resource: azurerm_linux_virtual_machine_scale_set.github_runners -> azurerm_kubernetes_cluster.inference xlabel: Deploy detail: After approval the runners roll the new model out to the online endpoint on AKS - resource: azurerm_kubernetes_cluster.inference -> azurerm_private_endpoint.blob xlabel: Load detail: AKS loads the model files from Blob Storage and the image from Container Registry - resource: tv_azurerm_users.claims_handlers -> azurerm_linux_web_app.claims xlabel: Claim detail: Claims handlers open a claim in the claims app - resource: azurerm_linux_web_app.claims -> azurerm_kubernetes_cluster.inference xlabel: Score detail: which calls the endpoint over the virtual network for a fraud score