mlflow-plugin-manager 0.1.0

PyPI package management integrated directly into MLflow's web interface

MLflow Plugin Manager

Welcome to the MLflow Plugin Manager, inspired by wbond's package manager for Sublime Text. This tool seamlessly integrates with MLflow, allowing you to install, update, and uninstall MLflow plugins directly from the MLflow web interface.

Features

• 🛠 Install new MLflow plugins with ease.
• 🔄 Update existing plugins to their latest versions.
• 🗑 Uninstall plugins you no longer need.
• 🌐 Direct integration into the MLflow web interface.

Preview

Installation

Clone this repository:

git clone [e-mail]:thijsdezoete/mlflow-plugin-manager.git

Navigate to the cloned directory and install the package:

cd mlflow-plugin-manager
pip install -e .

Usage

To run the MLflow server with the Plugin Manager, use the following command:

mlflow server --app-name plugin_manager

Once the server is running, navigate to the MLflow web interface. You'll find the Plugin Manager integrated and ready for use at:

http://localhost:5000/plugin-manager/

Configuration

Environment Variables

• PLUGIN_SERVER_URL: URL of the plugin metadata server
  • Default: https://api.mlflowplugins.com (production)
  • Local Development: Set to http://localhost:5001 if running your own server

Example for local development:

export PLUGIN_SERVER_URL="http://localhost:5001"
mlflow server --app-name plugin_manager

Architecture

The MLflow Plugin Manager consists of two components:

1. MLflow Plugin (this package): Integrates with MLflow's web interface and handles package installation locally via pip
2. Metadata Server: Provides plugin information (hosted at api.mlflowplugins.com)

Important: Package installation happens locally on your machine. The remote server only provides plugin metadata (available packages, versions, etc.).

For Server Maintainers Only

If you're maintaining your own plugin metadata server:

cd server
python reindex_plugins.py  # Index packages from PyPI
python app.py              # Run metadata server on port 5001

End users don't need to run these commands - the default configuration uses api.mlflowplugins.com.

License

This project is licensed under the MIT License. Refer to the LICENSE file for more details.