How to backup all MLFlow Tracking and Model Registry data from a Databricks Workspace

[alec]

Dear all,

I have a few questions, so please bear with me.

The MLFlow documentation does not provide any information regarding backups, except here. If I managed my own tracking server, I could make a backup, but how to I achieve this within Databricks?

pprint([dict(e)["artifact_location"] for e in MlflowClient().list_experiments()])

shows that the artifact locations are in dbfs:/databricks/mlflow-tracking/, but these are accessible only with the client ( https://docs.microsoft.com/de-de/azure/databricks/security/access-control/workspace-acl#mlflow-artifact-permissions ) I had thought about a blunt rsync copy all artifacts, but these would probably be problematic for the model lineage stored in the Registry.

I was hoping to get an idea of the tracking server inside a Databricks Notebook, with the env variable MLFLOW_TRACKING_URI, but its value is only:

MLFLOW_TRACKING_URI=databricks

3. In short, if I would delete a Databricks Workspace and wanted to have the full Tracking and Model Registry in a new one, how would I go about it?

Thank you for any help!

-Alec

[Alejandro Bolaños Rosales]

I found the answer here: https://github.com/amesar/mlflow-export-import

--
You received this message because you are subscribed to a topic in the Google Groups "mlflow-users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/mlflow-users/Kttrnxcw3C4/unsubscribe.
To unsubscribe from this group and all its topics, send an email to mlflow-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mlflow-users/0a05f049-aaa0-4058-ba7e-1542f59cd754n%40googlegroups.com.