Docker Compose Infrastructure

mlflow-experience provides a production-grade MLflow tracking server setup in compose.yaml.

---

1. What Services Run in Docker?

graph TD
    Client[Python Pipeline / Orchestrator] -->|Port 5000| MLflow[MLflow Server]
    MLflow -->|Metadata & Metrics| Postgres[(PostgreSQL 16 DB)]
    MLflow -->|Model & Log Artifacts| MinIO[(MinIO S3 Storage)]
    User[Developer] -->|Web UI Port 5000| MLflow
    User -->|Web Console Port 9001| MinIO

1. MLflow Server (mlflow):
2. Web UI & REST API running on port 5000.
3. Coordinates experiment runs, parameters, metrics, and models.
4. PostgreSQL Database (postgres):
5. Stores all relational metadata (experiment names, run IDs, tags, parameters, and metric histories).
6. Initialized using init-db.sql.
7. MinIO Object Storage (minio):
8. S3-compatible object storage for artifacts (e.g. trained model pickles, execution log files).
9. S3 API on port 9000, Web Console on port 9001.

---

2. Environment Settings (.env)

Service credentials and ports are managed in .env:

# PostgreSQL
POSTGRES_DB=mlflowdb
POSTGRES_USER=mlflow
POSTGRES_PASSWORD=mlflowpass

# MinIO
MINIO_ROOT_USER=minioadmin
MINIO_ROOT_PASSWORD=minioadminpass

# MLflow Server
MLFLOW_HOST=0.0.0.0
MLFLOW_PORT=5000

---

3. Starting the Services

Start in the background:

docker compose up -d

Check status:

docker compose ps

Stop the services:

docker compose down

---

4. Web Dashboards

• MLflow Tracking UI: http://localhost:5000
• MinIO Storage Console: http://localhost:9001
• Username: minioadmin
• Password: minioadminpass
• Bucket name: mlflow

---

5. Don't Have Docker?

If you don't have Docker installed or running on your system, you don't need it! You can start a local MLflow server directly with uv:

uv run mlflow server --backend-store-uri sqlite:///mlflow.db --default-artifact-root ./mlruns --port 5000