Azure deployment

Deploy GraphRAG with Azure OpenAI and Azure cloud services

This guide walks you through deploying GraphRAG using Azure services, including Azure OpenAI for language models and Azure Storage for scalable data management.

Why Azure?

Azure provides enterprise-grade features for GraphRAG deployments:

• Managed OpenAI - No API key rotation, enterprise SLAs
• Scalable storage - Blob Storage and Cosmos DB integration
• Security - Managed identities, VNet integration, private endpoints
• Compliance - Meet regulatory requirements with Azure’s certifications
• Cost management - Detailed billing and budget controls

Prerequisites

1. Azure subscription

   Ensure you have an active Azure subscription with appropriate permissions.

2. Azure CLI

   Install the Azure CLI:

   macOS

   brew install azure-cli
   
3. Login to Azure

   Authenticate with your Azure account:

   az login
   az account set --subscription "Your Subscription Name"
   

Set up Azure OpenAI

1. Create Azure OpenAI resource

   Create an Azure OpenAI service instance:

   az cognitiveservices account create \
     --name graphrag-openai \
     --resource-group graphrag-resources \
     --location eastus \
     --kind OpenAI \
     --sku S0 \
     --custom-domain graphrag-openai
   
2. Deploy models

   Deploy the required models (chat and embeddings):

   # Deploy GPT-4 for chat
   az cognitiveservices account deployment create \
     --name graphrag-openai \
     --resource-group graphrag-resources \
     --deployment-name gpt-4-deployment \
     --model-name gpt-4 \
     --model-version "0613" \
     --model-format OpenAI \
     --sku-capacity 10 \
     --sku-name "Standard"
   
   # Deploy text-embedding-3-small for embeddings
   az cognitiveservices account deployment create \
     --name graphrag-openai \
     --resource-group graphrag-resources \
     --deployment-name embedding-deployment \
     --model-name text-embedding-3-small \
     --model-version "1" \
     --model-format OpenAI \
     --sku-capacity 10 \
     --sku-name "Standard"
   
3. Retrieve endpoint and key

   Get your Azure OpenAI endpoint and API key:

   # Get endpoint
   az cognitiveservices account show \
     --name graphrag-openai \
     --resource-group graphrag-resources \
     --query "properties.endpoint" \
     --output tsv
   
   # Get API key
   az cognitiveservices account keys list \
     --name graphrag-openai \
     --resource-group graphrag-resources \
     --query "key1" \
     --output tsv
   

Configure Azure Storage

1. Create storage account

   Create an Azure Storage account for your data:

   az storage account create \
     --name graphragstorage \
     --resource-group graphrag-resources \
     --location eastus \
     --sku Standard_LRS \
     --kind StorageV2
   
2. Create containers

   Create containers for input and output data:

   # Get connection string
   CONNECTION_STRING=$(az storage account show-connection-string \
     --name graphragstorage \
     --resource-group graphrag-resources \
     --query "connectionString" \
     --output tsv)
   
   # Create containers
   az storage container create \
     --name graphrag-input \
     --connection-string "$CONNECTION_STRING"
   
   az storage container create \
     --name graphrag-output \
     --connection-string "$CONNECTION_STRING"
   

Configure GraphRAG for Azure

1. Update environment variables

   Create or update your .env file:

   .env

   GRAPHRAG_API_KEY=your-azure-openai-api-key
   AZURE_STORAGE_CONNECTION_STRING=your-storage-connection-string
   
2. Configure settings.yaml

   Update settings.yaml with Azure-specific configuration:

   settings.yaml

   # Azure OpenAI Configuration
   completion_models:
     default_completion_model:
       type: chat
       model_provider: azure
       model: gpt-4
       deployment_name: gpt-4-deployment
       api_base: https://graphrag-openai.openai.azure.com
       api_version: 2024-02-15-preview
       api_key: ${GRAPHRAG_API_KEY}
   
   embedding_models:
     default_embedding_model:
       type: embedding
       model_provider: azure
       model: text-embedding-3-small
       deployment_name: embedding-deployment
       api_base: https://graphrag-openai.openai.azure.com
       api_version: 2024-02-15-preview
       api_key: ${GRAPHRAG_API_KEY}
   
   # Azure Blob Storage for Input
   input:
     storage:
       type: blob
       connection_string: ${AZURE_STORAGE_CONNECTION_STRING}
       container_name: graphrag-input
     type: text
     file_pattern: .*\.txt$
   
   # Azure Blob Storage for Output
   output:
     type: blob
     connection_string: ${AZURE_STORAGE_CONNECTION_STRING}
     container_name: graphrag-output
   
   # Optional: Azure Blob Storage for Cache
   cache:
     type: json
     storage:
       type: blob
       connection_string: ${AZURE_STORAGE_CONNECTION_STRING}
       container_name: graphrag-cache
   

Using managed identity (recommended)

For production deployments, use Azure Managed Identity instead of API keys:

1. Create managed identity

   Create a user-assigned managed identity:

   az identity create \
     --name graphrag-identity \
     --resource-group graphrag-resources
   
2. Grant permissions

   Assign the managed identity to Azure OpenAI:

   # Get principal ID
   PRINCIPAL_ID=$(az identity show \
     --name graphrag-identity \
     --resource-group graphrag-resources \
     --query "principalId" \
     --output tsv)
   
   # Assign Cognitive Services User role
   az role assignment create \
     --role "Cognitive Services User" \
     --assignee "$PRINCIPAL_ID" \
     --scope "/subscriptions/YOUR_SUBSCRIPTION_ID/resourceGroups/graphrag-resources/providers/Microsoft.CognitiveServices/accounts/graphrag-openai"
   
   # Assign Storage Blob Data Contributor role
   az role assignment create \
     --role "Storage Blob Data Contributor" \
     --assignee "$PRINCIPAL_ID" \
     --scope "/subscriptions/YOUR_SUBSCRIPTION_ID/resourceGroups/graphrag-resources/providers/Microsoft.Storage/storageAccounts/graphragstorage"
   
3. Update configuration

   Modify settings.yaml to use managed identity:

   completion_models:
     default_completion_model:
       type: chat
       model_provider: azure
       model: gpt-4
       deployment_name: gpt-4-deployment
       api_base: https://graphrag-openai.openai.azure.com
       api_version: 2024-02-15-preview
       auth_method: azure_managed_identity  # Use managed identity
       # Remove api_key line
   
4. Authenticate Azure CLI

   Login with the managed identity:

   az login --identity
   

Deploy to Azure Container Instances

Run GraphRAG in Azure Container Instances for scheduled indexing:

1. Create Dockerfile

   Dockerfile

   FROM python:3.11-slim
   
   WORKDIR /app
   
   RUN pip install graphrag
   
   COPY settings.yaml .
   COPY .env .
   
   CMD ["graphrag", "index", "--root", "/app"]
   
2. Build and push image

   Build and push to Azure Container Registry:

   # Create container registry
   az acr create \
     --name graphragregistry \
     --resource-group graphrag-resources \
     --sku Basic
   
   # Build and push
   az acr build \
     --registry graphragregistry \
     --image graphrag:latest .
   
3. Deploy to ACI

   Create a container instance:

   az container create \
     --resource-group graphrag-resources \
     --name graphrag-indexer \
     --image graphragregistry.azurecr.io/graphrag:latest \
     --cpu 2 \
     --memory 4 \
     --registry-login-server graphragregistry.azurecr.io \
     --registry-username $(az acr credential show --name graphragregistry --query username -o tsv) \
     --registry-password $(az acr credential show --name graphragregistry --query passwords[0].value -o tsv) \
     --environment-variables \
       GRAPHRAG_API_KEY="$GRAPHRAG_API_KEY" \
       AZURE_STORAGE_CONNECTION_STRING="$CONNECTION_STRING"
   

Optional: Azure Cosmos DB storage

For enhanced scalability, use Azure Cosmos DB:

1. Create Cosmos DB account

   az cosmosdb create \
     --name graphrag-cosmos \
     --resource-group graphrag-resources \
     --kind GlobalDocumentDB
   
2. Configure in settings.yaml

   output:
     type: cosmosdb
     connection_string: ${COSMOS_CONNECTION_STRING}
     database_name: graphrag
     container_name: output
   

Cost optimization

Model selection

Choose cost-effective models:

• Use gpt-3.5-turbo instead of gpt-4 for initial testing
• Use text-embedding-3-small instead of text-embedding-3-large

Rate limiting

Configure rate limits to control costs:

completion_models:
  default_completion_model:
    rate_limit:
      requests_per_period: 60
      period_in_seconds: 60

Storage tiers

Use appropriate storage tiers:

• Hot tier for active data
• Cool tier for archival
• Configure lifecycle policies

Provisioned throughput

For Azure OpenAI:

• Use provisioned throughput for predictable workloads
• Monitor and adjust capacity

Monitoring and logging

1. Enable diagnostics

   Enable diagnostic logging for Azure OpenAI:

   az monitor diagnostic-settings create \
     --name graphrag-diagnostics \
     --resource "/subscriptions/YOUR_SUBSCRIPTION_ID/resourceGroups/graphrag-resources/providers/Microsoft.CognitiveServices/accounts/graphrag-openai" \
     --logs '[{"category": "RequestResponse", "enabled": true}]' \
     --metrics '[{"category": "AllMetrics", "enabled": true}]' \
     --workspace YOUR_LOG_ANALYTICS_WORKSPACE_ID
   
2. Set up alerts

   Create alerts for cost and performance:

   az monitor metrics alert create \
     --name high-token-usage \
     --resource-group graphrag-resources \
     --scopes "/subscriptions/YOUR_SUBSCRIPTION_ID/resourceGroups/graphrag-resources/providers/Microsoft.CognitiveServices/accounts/graphrag-openai" \
     --condition "total ProcessedPromptTokens > 1000000" \
     --description "Alert when token usage exceeds threshold"
   

Security best practices

Use managed identities

Avoid storing credentials, use Azure Managed Identity

Private endpoints

Configure private endpoints for Azure services

Network security

Implement VNet integration and firewall rules

Key rotation

Automate API key rotation using Key Vault

Next steps

Multi-lingual support

Deploy GraphRAG for multiple languages

Enterprise knowledge

Enterprise deployment patterns

Configuration reference

Complete configuration guide

Azure documentation

Azure OpenAI documentation