AWS Certified Machine Learning – SpecialtyMachine Learning Implementation and OperationsMedium
A data science team is developing a critical machine learning model for real-time anomaly detection. The model needs to be updated frequently, but the team wants to minimize downtime and risk during deployments. They also need a mechanism to quickly roll back to the previous stable version if issues are detected with the new model. Which deployment strategy is most suitable for this scenario?
- AA/B Testing
- BIn-Place Update
- CBlue/Green Deployment
- DRolling Update
Show answer & explanationAnswer & explanation
Correct answer: C. Blue/Green Deployment
Blue/Green deployment involves running two identical environments, 'Blue' (current stable) and 'Green' (new version). Traffic is gradually shifted to 'Green'. If issues arise, traffic can be instantly reverted to 'Blue', minimizing downtime and enabling quick rollbacks, which aligns perfectly with the requirements for critical models and frequent updates.
Why the other options are wrong
- A. A/B testing is for comparing model performance, not primarily for minimizing deployment risk or enabling quick rollbacks.
- B. In-place updates directly modify the running environment, leading to downtime and high risk during deployment.
- D. Rolling updates update instances one by one, which can still experience issues during the transition and doesn't allow instant rollback of all traffic.
Blue/Green Deployment (ML)
A deployment strategy where two identical production environments (Blue: current, Green: new) are maintained. Traffic is shifted from Blue to Green, allowing for quick rollback if issues occur.
- Minimizes downtime during deployment.
- Enables instant rollback to the previous stable version.
- Requires double the infrastructure capacity temporarily.
Memory trick: Blue/Green is like switching traffic lights: safe and fast.