AWS Lambda to Azure Functions Migration

Detailed guidance for migrating AWS Lambda functions to Azure Functions.

Overview

Programming Model Mapping

Trigger Mapping

Runtime-Specific Migration Patterns

For language-specific migration rules, correct/incorrect patterns, and code examples, see the runtime reference for the target language:

Project Structure

REQUIRED for Azure Functions:
src/
├── app.js (or function_app.py)   # Main entry point
├── host.json                      # Function host configuration
├── local.settings.json            # Local development settings
├── package.json (or requirements.txt)
├── [helper-modules]               # Business logic
└── tests/                         # Test files

❌ NEVER create:
├── [functionName]/                # No individual function directories
│   ├── function.json              # No function.json (JS v4, Python v2)
│   └── index.js

Environment Variables

✅ Use managed identity connections for Azure Functions storage:
   AzureWebJobsStorage__blobServiceUri
   AzureWebJobsStorage__queueServiceUri
   AzureWebJobsStorage__tableServiceUri

✅ Use specific endpoint variables for other services:
   COMPUTER_VISION_ENDPOINT
   STORAGE_ACCOUNT_URL
   SOURCE_CONTAINER_NAME

❌ Avoid:
   CONNECTION_STRING (use managed identity)
   API_KEY (use managed identity)

Reference Links

• AWS Lambda vs Azure Functions comparison
• AWS to Azure services comparison
• Supported language runtimes
• Triggers and bindings overview
• Functions quickstart JS (azd)
• Functions quickstart .NET Event Grid (azd)

Flex Consumption + Blob Trigger with EventGrid Source

⚠️ CRITICAL: When deploying blob triggers with source: 'EventGrid' on Flex Consumption, there are three infrastructure requirements that are NOT automatically handled and will cause silent trigger failures if missed.

1. Always-Ready Instances (Bootstrap Problem)

On Flex Consumption, trigger groups only start when there's work to do. But the blob extension needs to be running to create the Event Grid subscription that would deliver work — a chicken-and-egg problem.

Solution: Configure alwaysReady for the blob trigger group in the function app's scaleAndConcurrency:

// In api.bicep — functionAppConfig section
scaleAndConcurrency: {
  alwaysReady: [
    {
      name: 'blob'
      instanceCount: 1
    }
  ]
  instanceMemoryMB: 2048
  maximumInstanceCount: 100
}

Without this, the trigger group never starts → Event Grid subscription never gets created → no events are delivered → function never triggers.

2. Queue Endpoint Required

The blob extension internally uses Storage Queues for poison-message tracking when source: 'EventGrid' is configured. Without the queue endpoint, the function fails to index with:

Unable to find matching constructor while trying to create an instance of QueueServiceClient.
Expected: serviceUri. Found: credential, clientId, blobServiceUri

Solution: Always enable the queue endpoint alongside blob when using EventGrid source:

// In identity-based storage configuration
AzureWebJobsStorage__blobServiceUri: storageAccount.properties.primaryEndpoints.blob
AzureWebJobsStorage__queueServiceUri: storageAccount.properties.primaryEndpoints.queue  // REQUIRED for EventGrid source
AzureWebJobsStorage__credential: 'managedidentity'
AzureWebJobsStorage__clientId: managedIdentityClientId

Also assign Storage Queue Data Contributor RBAC role to the UAMI.

3. Event Grid Subscription via Bicep (Not CLI)

Do NOT create Event Grid event subscriptions via CLI. The az eventgrid system-topic event-subscription create command requires a webhook validation handshake that consistently fails on Flex Consumption with "response code Unknown" (timeout during cold start).

Solution: Deploy the Event Grid system topic and event subscription as Bicep resources. ARM handles the webhook validation internally and reliably:

// eventGrid.bicep
resource systemTopic 'Microsoft.EventGrid/systemTopics@2024-06-01-preview' = {
  name: 'evgt-${storageAccountName}'
  location: location
  properties: {
    source: storageAccount.id
    topicType: 'Microsoft.Storage.StorageAccounts'
  }
}

resource eventSubscription 'Microsoft.EventGrid/systemTopics/eventSubscriptions@2024-06-01-preview' = {
  parent: systemTopic
  name: 'blob-trigger-sub'
  properties: {
    destination: {
      endpointType: 'WebHook'
      properties: {
        // ARM resolves system key and handles validation at deployment time
        endpointUrl: 'https://${functionApp.properties.defaultHostName}/runtime/webhooks/blobs?functionName=${functionName}&code=${listKeys('${functionApp.id}/host/default', '2023-12-01').systemKeys.blobs_extension}'
      }
    }
    filter: {
      includedEventTypes: [ 'Microsoft.Storage.BlobCreated' ]
      subjectBeginsWith: '/blobServices/default/containers/${sourceContainerName}/'
    }
  }
}

RBAC requirement: Assign EventGrid EventSubscription Contributor role to the UAMI.

User Assigned Managed Identity (UAMI) Auth Patterns

DefaultAzureCredential with UAMI

When using UAMI, DefaultAzureCredential() without arguments tries SystemAssigned first and fails. Always pass the client ID:

const { DefaultAzureCredential } = require('@azure/identity');

const credential = new DefaultAzureCredential({
  managedIdentityClientId: process.env.AZURE_CLIENT_ID
});

Add AZURE_CLIENT_ID as an app setting pointing to the UAMI client ID:

appSettings: {
  AZURE_CLIENT_ID: managedIdentity.outputs.clientId
}

Identity-Linked Storage Connection

AzureWebJobsStorage__credential: 'managedidentity'
AzureWebJobsStorage__clientId: managedIdentityClientId
AzureWebJobsStorage__blobServiceUri: storageAccount.properties.primaryEndpoints.blob
AzureWebJobsStorage__queueServiceUri: storageAccount.properties.primaryEndpoints.queue

Required RBAC Roles for Face Blur Pattern

AWS Rekognition → Azure AI Computer Vision

⚠️ Package version: @azure-rest/ai-vision-image-analysis is still in beta. The ^1.0.0 semver does NOT resolve. Pin explicitly: "@azure-rest/ai-vision-image-analysis": "1.0.0-beta.3"

Coordinate Conversion

AWS Rekognition returns relative coordinates (0-1). Azure AI returns pixel coordinates. Convert for consistent face processing:

const sharp = require('sharp');
const metadata = await sharp(imageBuffer).metadata();

const faces = result.body.peopleResult.values.map(person => ({
  BoundingBox: {
    Width: person.boundingBox.width / metadata.width,
    Height: person.boundingBox.height / metadata.height,
    Left: person.boundingBox.x / metadata.width,
    Top: person.boundingBox.y / metadata.height
  }
}));

Auth: Use UAMI (No API Keys)

// computerVision.bicep
resource computerVision 'Microsoft.CognitiveServices/accounts@2024-10-01' = {
  kind: 'ComputerVision'
  properties: {
    disableLocalAuth: true  // Enterprise policy compliance — no API keys
  }
}