Deployment strategies for updating ML models with SageMaker Endpoint Production Variants
In this section, we will dive into multiple deployment strategies you can adopt to update production models using SageMaker Endpoint Production Variants. While some deployment strategies are easy to implement and are cost-effective, others add complexity while lowering deployment risks. We will dive into five different strategies, including Standard, A/B, Blue/Green, Canary, and Shadow deployments, and discuss the various steps involved in each approach.
Standard deployment
This strategy is the most straightforward approach to deploying and updating models in production. In a Standard model deployment, there is always a single active SageMaker endpoint, and the endpoint is configured with a single production variant, which means only a single model is deployed behind the endpoint. All inference traffic is processed by a single model. The endpoint configuration is similar to Endpoint Configuration...
