AI-300 Practice Question 91: ACR Image Build Compute
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 10 · question 1 of 10
A workspace's Azure Container Registry is behind a private endpoint, and environment image builds fail because the default build compute cannot reach the registry. The security team does not allow public access to the registry. What should the ML engineer do?
Show answer and explanation
Answer:
By default, the workspace builds environment images on serverless compute. That default works only when the workspace's dependent resources, such as the registry, are not behind private endpoints. With the registry in a virtual network, the workspace's image_build_compute must be set to a CPU-only compute cluster in a network that can reach the registry, which then builds the images (option C).
- A is wrong because this removes the network isolation that security requires, so it is not an acceptable fix. - B is wrong because the failure is that the default build compute cannot reach the registry behind the private endpoint. Permissions are not the cause. - D is wrong because the region has no effect on whether the default build compute can reach a registry behind a private endpoint.
Reference: Secure an Azure Machine Learning workspace with virtual networks