MLFlow Service Connector - Ability to authenticate with MLFLOW_TRACKING_TOKEN OS env varable

[hakan-77]

Issue: MLFlow connector currently does not support any form of authentication and non-secured MLFLow deployments are rare for production deployments.

Solution: Ability to define MLFLOW_TRACKING_TOKEN OS env variable before connecting MLFLow. see details in docs

Implementation: I see the below code returns an MlflowClient instance. Defining (or updating) MLFLOW_TRACKING_TOKEN OS env variable one line above this would work since MLFLow support changing token on the fly to access multiple MLFlow deployments. (see details: mlflow/mlflow#8823)

Current code:

OpenMetadata/ingestion/src/metadata/ingestion/source/mlmodel/mlflow/connection.py

Line 37 in a3e5190

return MlflowClient(

Note: This change might also resolve #15405 since (to my knowledge) Databricks hosted MLFlow also support using MLFLOW_TRACKING_TOKEN. As well as other hosted MLFlow solutions from AWS Sagemaker etc.

[harshach]

@hakan-77 what authentication /sso integration do you use

[hakan-77]

@harshach we support multiple, Azure, Keycloak, Auth0... MLFlow does not really care how your OAuth flow works. Once the token is obtained, it needs to be added to the MlFlow's request Auth Bearer token. Setting MLFLOW_TRACKING_TOKEN OS env varaible does just that.