Skip to main content

Architecture

System Architecture Overview​

MLflow 3.0.1 implements a distributed architecture designed for enterprise-scale machine learning lifecycle management.

Core Components​

  • Tracking Server: Centralized REST API server for experiment tracking

  • Model Registry: Centralized model store with versioning and stage management

  • Artifact Store: Distributed storage for models, datasets, and experiment artifacts

  • Backend Store: Metadata database for tracking runs, experiments, and model registry

  • UI Server: Web-based interface for experiment visualization and model management

Architecture Layers

┌─────────────────────────────────────────────────────────────┐
│                    MLflow UI Layer                          │
├─────────────────────────────────────────────────────────────┤
│                    MLflow API Layer                         │
├─────────────────────────────────────────────────────────────┤
│   Tracking Server  │  Model Registry  │  Artifact Store    │
├─────────────────────────────────────────────────────────────┤
│              Backend Database (PostgreSQL)                  │
├─────────────────────────────────────────────────────────────┤
│                 Storage Layer (S3/GCS/Azure)               │
└─────────────────────────────────────────────────────────────┘

Network Architecture

  • Port 5000: MLflow Tracking Server

  • Port 5001: MLflow Model Registry

  • Port 8080: MLflow UI Server

  • Database: PostgreSQL on port 5432

  • Object Storage: S3-compatible storage