Summary
In this chapter, we reviewed the reasons we should update production ML models. You learned how to use Production Variants to host multiple models using a single SageMaker Endpoint. You then learned about multiple deployment strategies that balance the cost and risk of model updates with ease of implementation and rollbacks. You also learned about the various steps involved and the configurations to use for Standard, A/B, Blue/Green, Canary, and Shadow deployments.
This chapter concluded with a comparison of the pros and cons and the applicability of each deployment strategy to specific use cases. Using this discussion as guidance, you can now choose an appropriate strategy to update your production models so that they meet your model availability and model quality requirements.
In the next chapter, we will continue our discussion of deploying models and learn about optimizing model hosting and infrastructure costs.
