Blue/green deployments

In a production environment where our models are running to make inferences in real time or near real time, it is very important that when we need to update our endpoints that it can happen without any disruption or problems. Amazon SageMaker automatically uses blue/green deployment methodology whenever we update our endpoints. In this kind of scenario, a new fleet, called the green fleet, is provisioned with our updated endpoints. The workload is then shifted from the old fleet, called the blue fleet, to the green fleet. After an evaluation period to make sure that everything is running without any issues, the blue fleet is terminated. SageMaker also provides the following three different traffic-shifted modes for blue/green deployment, allowing us to have more control over the traffic-shifting patterns.

All at once

In this traffic-shifting mode, all of the traffic is shifted at once from the blue fleet to the green fleet. The blue (old) fleet is kept...