[SOLVED] Difference between tracking_uri and the backend store uri in MLFLOW

Difference between tracking_uri and the backend store uri in MLFLOW

I am using Mlflow for my project hosting it in an EC2 instance. I was wondering in MlFlow what is the difference between the backend_store_uri we set when we launch the server and the trarcking_uri ?

Thanks,

Solution

tracking_uri is the URL of the MLflow server (remote, or built-in in Databricks) that will be used to log metadata & model (see doc). In your case, this will be the URL pointing to your EC2 instance that should be configured in programs that will log parameters into your server.

backend_store_uri - is used by MLflow server to configure where to store this data - on filesystem, in SQL-compatible database, etc. (see doc). If you use SQL database, then you also need to provide the --default-artifact-root option to point where to store generated artifacts (images, model files, etc.)