azureml-mlflow 'log_artifact' throws a ClientAuthenticationError if the upload directory is malformed

[ocramz-mhc]

• Package Name: azureml-mlflow azure-ai-ml
• Package Version: azureml-mlflow==1.60.0 , azure-ai-ml==1.26.2
• Operating System: Linux
• Python Version: 3.10

Describe the bug

The MLFlow client exposed by azureml-mlflow throws an authentication exception when uploading an artifact to a MLFlow run, if the upload directory (artifact_path) is malformed (e.g. it is the '.' path ).

I suspect the exception is thrown by the underlying object storage SDK, but the AzureML sdk frontend should do some input validation IMO.

To Reproduce

from mlflow import log_artifact

log_artifact(local_path = 'whatever.jpg', 
      artifact_path = '.')

Expected behavior
The client should say that the upload directory is malformed/unexpected, rather than throwing an authentication exception.

Add some client-side input validation, please.

Screenshots

[PratibhaShrivastav18]

Thanks for reporting @ocramz-mhc. We are looking into this.

[PratibhaShrivastav18]

@ocramz-mhc I did try out the code snippet you've shared to repro the issue, more specifically below :

from azure.ai.ml import MLClient
from azure.identity import DefaultAzureCredential
import mlflow
from mlflow import log_artifact

# Initialize Azure ML client with explicit parameters
credential = DefaultAzureCredential()
subscription_id = ""
resource_group = ""
workspace_name = ""

ml_client = MLClient(
    credential=credential,
    subscription_id=subscription_id,
    resource_group_name=resource_group,
    workspace_name=workspace_name
)
mlflow_tracking_uri = ml_client.workspaces.get(ml_client.workspace_name).mlflow_tracking_uri
mlflow.set_tracking_uri(mlflow_tracking_uri)

mlflow.set_experiment("/my-experiment")

log_artifact(local_path='package.json', 
            artifact_path='.')

And it did upload the file successfully in the run. As for client validations, we're not currently limiting any paths, and it looks like the storage service takes care of normalising any invalid paths and tries to upload. Exceptions encountered, if any, are thrown by the storage service.

In your case, it looks like the actual issue is the Authentication exception you're seeing. Please feel free to add any other observations you have.