How to connect to MLFlow tracking server that has auth?

Question

I want to connect to remote tracking server (http://123.456.78.90) that requires authentication

When I do this:

import mlflow
mlflow.set_tracking_uri("http://123.456.78.90")
mlflow.set_experiment("my-experiment")

I get an error

MlflowException: API request to endpoint /api/2.0/mlflow/experiments/list failed with error code 401 != 200. Response body: 401 Authorization Required

I understand that I need to log in first but I have no idea how to do it

Schofield

Answer 1

MLflow documentation says:

MLFLOW_TRACKING_USERNAME and MLFLOW_TRACKING_PASSWORD - username and password to use with HTTP Basic authentication. To use Basic authentication, you must set both environment variables.

So you just need to set these variables in your code using os.environ:

os.environ['MLFLOW_TRACKING_USERNAME'] = 'name'
os.environ['MLFLOW_TRACKING_PASSWORD'] = 'pass'

Soapstone

Thanks, it worked! One more question, if I want to store artifacts in s3, what should I set as a tracking URL? mlflow or aws? – Schofield

It's really should be configure on the MLflow server: mlflow.org/docs/latest/… – Soapstone

Getting MlflowException: API request to endpoint /api/2.0/mlflow/experiments/list failed with error code 403 != 200 – Duda

Answer 2

os.environ['MLFLOW_TRACKING_USERNAME'] = 'name'
os.environ['MLFLOW_TRACKING_PASSWORD'] = 'pass'
os.environ['MLFLOW_TRACKING_URI'] = 'tracking_uri'

Amphichroic

Answer 3

Late to the party, but there is another way (see the MLflow authentication docs) of setting credentials globally (per system user), in the ~/.mlflow/credentials file (in INI format):

[mlflow]
mlflow_tracking_username = username
mlflow_tracking_password = password

This way, you don't have to set the environment variables in each script or your shell config. Any credentials defined in environment variables will override those from the credentials file.

Haze