MLflow Auth GUI

This web application provides a Web GUI to the MLflow Authentication API.

GUI provides:

• self-registration of users in MLflow
• self-registration service requires Authentication via a pre-configured OIDC Provider (e.g. EGI Check-In)
• one can limit users eligible for self-registration using REQUIRED_ENTITLEMENT
• created in the MLflow user name is the email address retrieved from the OIDC token (registered email)
• optionally, the MLflow user credentials are stored in the HashiCorp Vault Secret storage via AI4OS Platform API (ai4-papi)
• after the user is registered in MLflow, he/she can:
  • update his/her MLflow password
  • update permissions to his/her experiments
  • update permissions to his/her registered models
  • delete account

Deployment

The deployment is containerised and includes provision of the fully running MLflow service:

• GUI for the MLflow Authentication REST API (aka signup service)
• MLflow instance (mlflow service)
• Postgresql database as the backend store (database service)
• databases backup (backup_db service)
• traefik reverse proxy (reverse-proxy service)

docker compose configuration is based on awesome-compose.

Configuration

Copy .env.sample to .env and customize the values for your deployment.

Ensure MLFLOW_USERNAME & MLFLOW_PASSWORD in .env and admin_username & admin_password in backend/srv/auth_config.ini match.

OR

Once deployed, use e.g. mlflow_auth scripts to update MLflow admin user to match MLFLOW_USERNAME & MLFLOW_PASSWORD.

If setting up on windows, launch the compose once to create volumes & files, then uncomment user: postgres in compose.yml.

Deploy for development

1. Use docker compose up -d, docker compose overrides automatically using compose.override.yml file.
2. In signup:
   1. Install dependencies: yarn install
   2. Verify the env in .env, override it in .env.local if necessary
   3. Run yarn dev to start the development server

Deploy in Production

Use:

docker compose -f compose.yml -f compose.prod.yml up -d

Usage

For the example usage, please, see https://docs.ai4os.eu/en/latest/user/howto/mlops/mlflow.html

Contributing

Please, see our CONTRIBUTING description and the CODE OF CONDUCT.

License

This code is distributed under the Apache 2.0 License. Please, see the LICENSE file.

Copyright (c) 2023 - 2024 Karlsruhe Institute of Technology - Scientific Computing Center.