Connect Azure Synapse directly to Model Registry

Michael Wittman 76 Reputation points
2024-06-10T15:37:05.6+00:00

I'm attempting to connect my Azure Synapse Workspace to an Azure Machine Learning Registry resource directly, without the addition of a Machine Learning Workspace. I had previously had networking issues utilizing the workspace and, seeing as Synapse is going to be the main development platform, setting up a workspace felt redundant. I was able to create just a registry within the AZ ML portal, and am hoping we can use just the registry as our MLFlow server and model repository.

We are utilizing fully secured resources behind a VNET and connect everything via private endpoints. I have created PEs in the portal for synapse to connect to the model registry and the container registry resources, and created an MPE in the Synapse workspace to connect to the storage account under the model registry.

Within a Synapse notebook, I attempt to set the MLFlow tracking URI to the same URI present on the Model Registry's page. However, when running the notebook and attempting to set the experiment, I receive an "MlflowException: API request to endpoint /api/2.0/mlflow/experiments/create failed with error code 404 != 200. Response body: ''" error. I'm not seeing any way to create MPEs to either the container or model registries within the Synapse portal.

I'm not sure where I might be going wrong, but is it even possible to connect to the registry directly like this within a synapse environment?

Azure Machine Learning
Azure Machine Learning
An Azure machine learning service for building and deploying models.
2,723 questions
{count} votes