App service machine learning landing zone support

[hudua]

Overview/Summary

Added optional support for App Service (Linux container) in machine learning landing zone, so that model deployment can be to app service, instead of AKS. This way Azure ML can be used for model training and push a containerized model to Azure Container Registry with private endpoint. Then App Service (Linux container) can pull from the registry and run the model as a service.

This resolves issue #93

This PR fixes/adds/changes/removes

1. Machine learning landing zone documentation updated to reflect addition of app service
2. Machine learning landing zone bicep scripts updated to support app service as well as make AKS cluster optional
3. Updated schema definition and change log
4. Updated existing unit tests
5. Updated existing deployment JSON files
6. Ran schema tests/schemas/run-tests.sh test
7. Added some ad-hoc deployment tests for app service under tests/landingzones/lz-machinelearning/deployment-tests

Breaking Changes

Backwards compabilities may be affected per:

1. Introduced enabled field for AKS config file
2. Introduced additional integration subnet due to app service vnet integration delegation requirement

Testing Evidence

App service deployed:

Using Linux publicly available container: mcr.microsoft.com/appsvc/staticsite:latest

Virtual network integration set

As part of this Pull Request I have

• Checked for duplicate Pull Requests
• Associated it with relevant GitHub Issues
• Ensured my code/branch is up-to-date with the latest changes in the main branch
• Performed testing and provided evidence.
• Updated relevant and associated documentation.

[hudua]

A few pending work items should be done prior to approval, but require review first:

• Kindly ask @SenthuranSivananthan to run e2e deployment so I can verify using integration tests from Azure ML to ACR to App Service
• Kindly ask @Adeelku to review aks service deployment and schema object now that the enabled field is required, making AKS cluster an optional deployable service in ML landing zone
• Once testing is complete, descriptions and scripts on integration tests using App Service and Azure ML for the ML landing zone should be updated

In addition, App Service could have access restrictions and private endpoint for inbound traffic. Though for this pull request I'm leaving it out of scope and would suggest it be addressed in a separate issue

[SenthuranSivananthan]

App Service as a the model runtime is a great alternative. A few suggested changes so that the parameters are more explicit on the intent.

[SenthuranSivananthan]

I've left comments on small fixes to the templates.

Can you please add test cases for when aks.enabled = false and appService.enabled = false. We can then add these test cases as deployment scenarios in docs/archetypes/machinelearning.md#deployment-scenarios

[SenthuranSivananthan]

I've successfully tested the following 3 deployment scenarios:

• AKS & App Service Disabled
• App Service Enabled
• AKS Enabled

[hudua]

Thanks! Verified integration tests for Azure ML to App Service

Model package successful

Model pushed to ACR

App Service deployment

[SenthuranSivananthan]

Thank for the enhancement & additional deployment validation @hudua.

#approved