[FEATURE] Migrate AzureML service 'ContainerImage' usage to 'Environments'

[Chris113113]

Description

Creating and deploying a ContainerImage is soon-to-be deprecated functionality inside of AzureML service.

The new recommended flow is instead to prepare an Environment object that contains your Pip/Conda requirements and simply include this in your InferenceConfiguration to be supplied in a Model.deploy() call.

This speeds up image build and service deployment time, and allows subsequent service deployments to leverage existing prepared Environment images.

Documentation on Environments can be found here: https://docs.microsoft.com/en-us/azure/machine-learning/service/how-to-use-environments

In which platform does it happen?

Azure Machine Learning service

How do we replicate the issue?

• Train your model
• Register a model with AzureML service
• Create a ContainerImage
• Deploy that ContainerImage as an AKS Webservice

Expected behavior (i.e. solution)

• Create a shared Environment for training and deploying models
• Train your model locally or on a remote compute in the given environment
• Register a model with AzureML service
• Deploy that model as an AKS Webservice

Other Comments

[yueguoguo]

@Chris113113 Do we have an estimation about timeline for such deprecation?