Create an automated CI/CD pipeline using GitHub Actions that:

• Trains and registers a machine learning model
• Deploys it to a managed Azure endpoint
• Enables full observability: logging, alerts, and diagnostics

Provision these with Bicep or Terraform:

1. Azure ML Workspace
2. Azure Key Vault (for secrets like storage keys)
3. Azure Storage Account (for data input/output)
4. Azure Container Registry (optional: custom container inference)
5. Azure Application Insights (for logs and metrics)
6. Azure Monitor Alert Rules (trigger on failed jobs or degraded endpoints)
7. Compute cluster (for training, e.g., cpu-cluster)
8. Azure ML Online Endpoint (for model deployment)

plaintextCopyEdit.github/workflows/
├── train-deploy.yml         # GitHub Actions workflow
infra/
├── main.bicep               # Infrastructure as code
ml/
├── train.py                 # Model training script
├── score.py                 # Inference entry point
├── environment.yml          # Conda environment for training/deployment
├── register_model.py        # Registers trained model
├── pipeline_job.yml         # Azure ML pipeline definition (optional)

1. Checkout & Install Dependencies
2. Login to Azure (azure/login GitHub Action)
3. Set up Azure ML CLI or Python SDK
4. Run Training Script (optionally via pipeline YAML)
5. Register Model to Azure ML Registry
6. Deploy Model to Online Endpoint
7. Post-deployment Tests
8. Publish Logs to Application Insights
9. Trigger Alerts if any step fails (via az monitor alert rules)

• Application Insights: attach to the Azure ML endpoint for request/response logs and metrics.
• Azure Monitor Alerts:
  • Alert on failed training runs (via log analytics query).
  • Alert on high latency or low success rate on the deployed endpoint.
  • Notification via email/webhook/Teams.

1. Dashboards: Include an Azure Dashboard that surfaces training job status, endpoint performance, recent alerts.
2. Web Frontend (Optional): Simple app to send inference requests, visualize logs.
3. Cost Control: Auto-shutdown training compute after use.

Business Case: Retrain a churn prediction model every week using new customer data.

• GitHub Actions scheduled trigger: weekly
• Logs retraining results
• Deploys model if metrics (e.g., accuracy > previous version) pass
• Sends alerts if model accuracy drops >10% or job fails

• Use GitHub Secrets for Azure credentials
• Leverage Workload Identity Federation for GitHub-Azure auth
• RBAC for least-privilege access to ML and monitoring resources

• CI/CD pipeline runs end-to-end on commit
• Azure infrastructure deployed from code
• Model available at a public or private endpoint
• Logs visible in App Insights
• Alerts trigger on defined failure conditions

main.bicep

param location string = resourceGroup().location
param workspaceName string
param storageAccountName string
param keyVaultName string
param appInsightsName string
param logAnalyticsName string
param computeName string = 'cpu-cluster'

resource logAnalytics 'Microsoft.OperationalInsights/workspaces@2021-06-01' = {
  name: logAnalyticsName
  location: location
  properties: {
    retentionInDays: 30
    sku: {
      name: 'PerGB2018'
    }
  }
}

resource appInsights 'Microsoft.Insights/components@2020-02-02' = {
  name: appInsightsName
  location: location
  kind: 'web'
  properties: {
    Application_Type: 'web'
    WorkspaceResourceId: logAnalytics.id
  }
}

resource storage 'Microsoft.Storage/storageAccounts@2023-01-01' = {
  name: storageAccountName
  location: location
  sku: {
    name: 'Standard_LRS'
  }
  kind: 'StorageV2'
  properties: {}
}

resource keyVault 'Microsoft.KeyVault/vaults@2023-02-01' = {
  name: keyVaultName
  location: location
  properties: {
    tenantId: subscription().tenantId
    sku: {
      name: 'standard'
      family: 'A'
    }
    accessPolicies: []
    enabledForDeployment: true
    enabledForTemplateDeployment: true
    enableSoftDelete: true
  }
}

resource mlWorkspace 'Microsoft.MachineLearningServices/workspaces@2023-04-01' = {
  name: workspaceName
  location: location
  identity: {
    type: 'SystemAssigned'
  }
  properties: {
    description: 'Demo ML workspace for CI/CD'
    storageAccount: storage.id
    applicationInsights: appInsights.id
    keyVault: keyVault.id
    friendlyName: workspaceName
  }
}

resource computeCluster 'Microsoft.MachineLearningServices/workspaces/computes@2023-04-01' = {
  name: computeName
  parent: mlWorkspace
  location: location
  dependsOn: [mlWorkspace]
  properties: {
    computeType: 'AmlCompute'
    properties: {
      vmSize: 'STANDARD_DS3_V2'
      scaleSettings: {
        minNodeCount: 0
        maxNodeCount: 2
        idleTimeBeforeScaleDown: 'PT5M'
      }
    }
  }
}

Deployed with

az deployment group create \
  --name ml-cicd-deployment \
  --resource-group don-test-rg \
  --template-file main.bicep \
  --parameters workspaceName="don-ml-workspace" \
               storageAccountName="donmlstorage" \
               keyVaultName="donkv" \
               appInsightsName="donappinsights" \
               logAnalyticsName="donloganalytics"