notniknot/mlflow-docker

This repository contains a Docker-Setup for MLflow with integrated authentication and MinIO + PostgreSQL storage

This repository contains a Docker-Setup for MLflow with integrated authentication and MinIO + PostgreSQL storage

Addressed Issues • Target Group • Architecture • Setup • ToDos

Addressed Issues

This repro addresses the need for an easy solution to setup an MLflow-Docker-Environment.

Target Group

This repo is for data scientists and engineers who want to use MLflow with a solid storage and authentication solution.

Architecture

1. PostgreSQL database
   • Stores runs and metrics
   • Not accessible
2. Minio s3 storage
   • Stores models and artifacts
   • Accessible via Port 19000
3. MLflow server
   • Orchestrates all tasks
   • Accessible via Port 15000 (via ngninx)
4. nginx webserver
   • Acts as authentication layer for the mlflow server

Setup

Make sure the directory structure is as follows (you can check out my mlflow_extension as well that fits perfectly to this setup):

docker
├── docker-compose.yml
├── .env
├── minio
│   └── data
│       └── mlflow-bucket
├── mlflow_server
│   ├── Dockerfile
│   └── mlflow_extension-xxx-py3-none-any.whl (optional)
├── nginx
│   ├── Dockerfile
│   ├── .htpasswd
│   ├── mlflow.conf
│   └── nginx.conf
└── postgres
    └── data

• Rename .env-template file to .env
• Populate .env file with the proper values
• Decide whether you want to include the plugin or not
• Start with docker-compose -f "docker-compose.yml" up -d --build

ToDos

• Nothing yet