Correct:
* Create an environment.
Incorrect:
* Delete the Python 3.6 - AzureML kernel.
* Delete the Python 3.8 - AzureML kernel.
Note:
Before you can add a new Jupyter kernel on an Azure Machine Learning compute instance terminal, you must create a Conda environment.
Required Workflow
To officially provision and expose the new kernel to your Azure Machine Learning studio Notebooks, you need to execute the following full process from your terminal session:
Create the environment: Provision a new isolated environment (e.g., using conda create -n newenv python=3.10).
Activate the environment: Run conda activate newenv.
Install dependencies: Add the required ipykernel package using conda install ipykernel or pip install ipykernel.
Register the kernel: Bind the new environment configuration to the global Jupyter directory by running:
python -m ipykernel install --user --name newenv --display-name "My New Kernel" Reference:
https://docs.azure.cn/en-us/machine-learning/how-to-access-terminal

Batch endpoints are optimized for large-scale, asynchronous inference workloads. They efficiently process large datasets and scale based on demand, making them cost-effective for non-real-time scenarios. Online endpoints are designed for low-latency use cases and are more expensive for batch processing.

To restore the service as quickly as possible, you can roll back traffic to the previous deployment by updating the traffic allocation settings of your Azure Machine Learning online endpoint.
Azure Machine Learning managed online endpoints support multiple deployments under a single endpoint, allowing for blue-green deployment strategies where you can shift traffic between versions instantly.
Key Benefits of This Approach
Instant Recovery: Traffic shifting is a routing change and does not require redeploying the previous model's code or environment, making it the fastest recovery method.
No Downtime: Because the previous deployment remains "warm" (active but receiving no traffic), the switch happens without interrupting the service.
Isolation for Debugging: You can keep the failing deployment at 0% traffic to inspect its logs using az ml online-deployment get-logs without affecting end users How to Roll Back Traffic If your new deployment (e.g., "green") is returning errors, you can reallocate 100% of the traffic back to the known stable deployment (e.g., "blue") using the following methods:
* Azure CLI: Use the az ml online-endpoint update command to set the traffic percentage:
az ml online-endpoint update --name <your-endpoint-name> --traffic "blue=100 green=0"
* Azure Machine Learning Studio:
Navigate to Endpoints in the left menu.
Select your specific real-time endpoint.
Go to the Details or Live Traffic tab.
Adjust the traffic percentages so the previous deployment receives 100% and the failing deployment receives 0%.
Select Update or Save to apply the changes immediately
Reference:
https://learn.microsoft.com/en-us/answers/questions/1275110/azure-ml-v2-yaml-code-for-live-traffic-allocation

The ideal first step is to implement a Hub-and-Spoke Workspace Architecture combined with Azure Machine Learning Registries.
This hybrid blueprint natively decouples your organizational boundaries (billing and access) from your machine learning assets (models, environments, and pipelines).
In enterprise-scale machine learning, an Azure Machine Learning Hub Workspace acts as a central governance plane. It lets IT administrators configure security, shared compute resources, networks, and compliance guidelines centrally once. From this central hub, different teams can spin up isolated project workspaces. This provides strict data and team-level separation, which naturally supports precise cost tracking and billing isolation per team.
Reference:
https://learn.microsoft.com/en-us/azure/machine-learning/concept-hub-workspace