3.7. Deploy Containers for Local and Edge Devices

💡 First Principle: Some scenarios can't tolerate cloud latency or connectivity—factory floors, ships at sea, medical devices. Containers let you run Azure AI models locally, but Microsoft still needs to verify you're paying for usage. The billing endpoint is a "phone home" check that must succeed periodically (even if the actual inference runs offline). The exam loves testing this nuance: containers work offline for inference, but not for billing.

Building on the build-vs-buy framework from Section 1.4, containers offer a middle ground: use pre-built models but deploy them to your own infrastructure.

🔧 Implementation Reference: Containers

Container Deployment Pattern:

docker run --rm -it -p 5000:5000 \
  mcr.microsoft.com/azure-cognitive-services/textanalytics/language:latest \
  ApiKey={API_KEY} \
  Billing={ENDPOINT_URI} \
  Eula=accept

Available Container Services:

Docker Compose Example:

version: '3'
services:
  language:
    image: mcr.microsoft.com/azure-cognitive-services/textanalytics/language
    ports:
      - "5000:5000"
    environment:
      - ApiKey=${API_KEY}
      - Billing=${BILLING_ENDPOINT}
      - Eula=accept

⚠️ Exam Trap: Containers still need the Billing endpoint for metering—they don't work fully offline indefinitely.