I'm trying to avoid having to start an mlflow server to run an export, relying only on the mlflow python API to export from OSS MLFlow. I get an error similar to #56, Mlflow host or token is not configured correctly
The problem is that __init__.py in export_run.py tries to start a dbx REST client, which doesn't support MLFLOW_TRACKING_URI. The client is only ever used by the _export_notebook function.
Since I don't need notebook import/export functionality (and I don't believe this is supported by opensource mlflow), starting this client isn't necessary.
I've attached a basic PR which fixes the issue for me (I appreciate it may not be mergeable), but there are various other places in the codebase where a client is started in an init method and apparently only used for databricks import, e.g. in import_run.py.
ssabdb
commented
1 year ago
Hello, and thanks for this useful capability.
I'm trying to avoid having to start an mlflow server to run an export, relying only on the mlflow python API to export from OSS MLFlow. I get an error similar to #56,
Mlflow host or token is not configured correctlyThe problem is that __init__.py in export_run.py tries to start a dbx REST client, which doesn't support
MLFLOW_TRACKING_URI. The client is only ever used by the_export_notebookfunction.Since I don't need notebook import/export functionality (and I don't believe this is supported by opensource mlflow), starting this client isn't necessary.
I've attached a basic PR which fixes the issue for me (I appreciate it may not be mergeable), but there are various other places in the codebase where a client is started in an init method and apparently only used for databricks import, e.g. in
import_run.py.