Exam AI-300 Topic 1 Question 160 Discussion

To introduce a new model version in Azure Machine Learning without service interruption, you should use Blue/Green Deployment with Traffic Splitting.
This strategy allows you to run two versions of a model simultaneously under a single Online Endpoint, gradually shifting users to the new version once it is validated.
Key Benefits
Zero Downtime: The endpoint URL stays the same; only the backend routing changes.
Easy Rollback: If the new model fails, you can instantly flip traffic back to 100% on the old version.
Risk Mitigation: Only a small subset of users is exposed to the unproven model initially.
Implementation Steps
1. Create the "Green" Deployment
Deploy the new model version as a second deployment under the existing online endpoint.
Initially, set its traffic allocation to 0%.
2. Canary Testing (Initial Split)
Shift a small percentage of traffic (e.g., 10%) to the new deployment. Monitor performance metrics, error rates, and model accuracy in a real-world environment.
3. Validation & Monitoring
Use Azure Monitor and Application Insights to compare the two deployments. Check for:
Latency: Is the new model slower?
HTTP Status Codes: Are there 4xx or 5xx errors?
Model Drift: Is the prediction quality as expected?
4. Full Rollout
If the new version is stable, increase the traffic split (e.g., 50/50) until the new model handles
100% of the traffic.
5. Cleanup
Once the "Green" deployment is confirmed as the new production standard, you can delete the old ("Blue") deployment to save costs.
Reference:
https://learn.microsoft.com/en-us/azure/well-architected/ai/operations