Summary: Properly Hosting Azure Container Apps

In this module, you learned that deploying a container is just the beginning. The real challenge of AI cloud solutions is day-two operations: managing the lifecycle, tuning resources, and rapidly troubleshooting issues.

Here is the complete picture of how an Azure Container App is "properly hosted" in a production environment:

flowchart TD
    %% Define styles for Azure resources
    classDef acr fill:#0072C6,stroke:#0072C6,color:#fff,rx:5px
    classDef aca fill:#762C9D,stroke:#762C9D,color:#fff,rx:5px
    classDef logAnalytics fill:#68217A,stroke:#68217A,color:#fff,rx:5px
    classDef ingress fill:#FFB900,stroke:#FFB900,color:#333,rx:5px
    
    User([🌐 End Users])

    subgraph "Azure Container Apps Environment (Security Boundary)"
        Ingress[🚦 Managed Ingress / Load Balancer]:::ingress
        
        subgraph "AI API (Container App)"
            Rev1[📄 Old Revision<br>(Traffic: 0%)]:::aca
            Rev2[🚀 New Revision<br>(Traffic: 100%)]:::aca
        end
    end

    ACR[(🐳 Azure Container Registry<br>Images: amd64)]:::acr
    LogAnalytics[(📊 Log Analytics Workspace<br>Historical DVR)]:::logAnalytics

    %% Relationships
    User -- HTTPS --> Ingress
    Ingress -- Routes Traffic --> Rev2
    Ingress -. Shift Traffic .-> Rev1
    
    ACR -- Pulls Image --> Rev2
    
    Rev1 -- Streams Logs --> LogAnalytics
    Rev2 -- Streams Logs --> LogAnalytics

Key Takeaways for Proper Hosting

1. Architecture & Revisions: A properly hosted Container App doesn't just overwrite itself. Every configuration change or code update creates a new, immutable Revision. This enables zero-downtime deployments and instant rollbacks if a bad environment variable takes down the app.

2. Resource Optimization:

   • CPU: Think of it as the worker's processing speed. Starve it, and you get throttling (slow APIs).
   • RAM: Think of it as the desk space. Starve it, and you get Out Of Memory (OOM) crashes.

3. Health Probes: Never deploy without them.

   • Readiness Probes tell the load balancer, "Don't send me traffic yet, my AI models are still loading!"
   • Liveness Probes tell the platform, "I'm permanently stuck, please kill me and restart."

4. Debugging Mindset (Live TV vs. DVR):

   • Use az containerapp logs show for real-time, live debugging when actively developing.
   • Use Log Analytics with KQL (az monitor log-analytics query) to search historical logs, find errors that happened overnight, and analyze trends across multiple container restarts.

Cleanup Complete

The aca-lab06 resource group has been successfully submitted for deletion. It will disappear from your Azure account shortly, stopping any further billing.