Implementation:Mlflow Mlflow Docker Compose Deployment

Knowledge Sources	Mlflow_Mlflow
Domains	Deployment, Infrastructure
Last Updated	2026-02-13 20:00 GMT

Overview

A Docker Compose configuration that provides a one-command deployment of the MLflow tracking server with PostgreSQL as the backend store and MinIO as the S3-compatible artifact store.

Description

docker-compose.yml defines a multi-container deployment stack consisting of four services:

• postgres: A PostgreSQL 15 database that serves as the MLflow backend store for experiment and run metadata. It uses a named volume (pgdata) for persistent storage and includes a health check using pg_isready.
• minio: A MinIO object storage server that provides S3-compatible artifact storage. It exposes both the API port (9000) and the console port (9001), uses a named volume (minio-data) for persistence, and includes a health check via the MinIO health endpoint.
• create-bucket: An initialization service using the MinIO client (mc) that creates the artifact bucket after MinIO becomes healthy. This service runs once and does not restart.
• mlflow: The MLflow tracking server running from the official ghcr.io/mlflow/mlflow image. It depends on all other services being healthy or completed, installs psycopg2-binary and boto3 at startup, and launches mlflow server with the configured backend store URI and artifact destination with --serve-artifacts enabled.

All services communicate over a custom Docker network named mlflow-network. Configuration is externalized through environment variables, with defaults provided in the companion .env.dev.example file.

Usage

Use this configuration for local development, testing, or lightweight production deployments of MLflow with durable storage backends. Copy .env.dev.example to .env and customize the values before running.

Code Reference

Source Location

• Repository: Mlflow_Mlflow
• File: docker-compose/docker-compose.yml
• Lines: 1-109

Configuration

volumes:
  pgdata:
  minio-data:

services:
  postgres:
    image: postgres:15
    container_name: mlflow-postgres
    environment:
      POSTGRES_USER: ${POSTGRES_USER}
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD}
      POSTGRES_DB: ${POSTGRES_DB}
    volumes:
      - pgdata:/var/lib/postgresql/data
    ports:
      - "5432:5432"

  minio:
    image: minio/minio:latest
    container_name: mlflow-minio
    command: server /data --console-address ":9001"
    ports:
      - "9000:9000"
      - "9001:9001"

  create-bucket:
    image: minio/mc:latest
    container_name: mlflow-create-bucket
    depends_on:
      minio:
        condition: service_healthy

  mlflow:
    image: ghcr.io/mlflow/mlflow:${MLFLOW_VERSION}
    container_name: mlflow-server
    depends_on:
      postgres:
        condition: service_healthy
      minio:
        condition: service_healthy
      create-bucket:
        condition: service_completed_successfully
    ports:
      - "${MLFLOW_PORT}:${MLFLOW_PORT}"

networks:
  default:
    name: mlflow-network

I/O Contract

Inputs

Name	Type	Required	Description
POSTGRES_USER	env var	Yes	PostgreSQL username
POSTGRES_PASSWORD	env var	Yes	PostgreSQL password
POSTGRES_DB	env var	Yes	PostgreSQL database name
MINIO_ROOT_USER	env var	Yes	MinIO root user
MINIO_ROOT_PASSWORD	env var	Yes	MinIO root password
MINIO_HOST	env var	Yes	MinIO hostname (typically "minio" within the Docker network)
MINIO_PORT	env var	Yes	MinIO API port (default: 9000)
MINIO_BUCKET	env var	No	MinIO bucket name (default: "mlflow")
MLFLOW_BACKEND_STORE_URI	env var	Yes	PostgreSQL connection string for MLflow
MLFLOW_ARTIFACTS_DESTINATION	env var	Yes	S3 URI for artifact storage (e.g., "s3://mlflow/")
MLFLOW_S3_ENDPOINT_URL	env var	Yes	MinIO endpoint URL for S3 compatibility
AWS_DEFAULT_REGION	env var	Yes	AWS region for Boto3 (e.g., "us-east-1")
MLFLOW_HOST	env var	Yes	MLflow server bind address (e.g., "0.0.0.0")
MLFLOW_PORT	env var	Yes	MLflow server port (e.g., 5000)
MLFLOW_VERSION	env var	Yes	MLflow Docker image tag (e.g., "v3.3.0" or "latest")

Outputs

Name	Type	Description
MLflow UI	HTTP	MLflow tracking server accessible at http://localhost:${MLFLOW_PORT}
MinIO Console	HTTP	MinIO web console accessible at http://localhost:9001
PostgreSQL	TCP	PostgreSQL database accessible at localhost:5432

Usage Examples

Basic Startup

# Copy and customize environment variables
cd docker-compose
cp .env.dev.example .env

# Start all services
docker compose up -d

# Verify MLflow is running
curl http://localhost:5000/health

Tear Down

# Stop and remove containers (volumes persist)
docker compose down

# Stop and remove containers AND volumes
docker compose down -v

Service Dependency Graph

The services start in the following order based on health check dependencies:

1. postgres and minio start in parallel
2. create-bucket starts after minio is healthy
3. mlflow starts after postgres is healthy, minio is healthy, and create-bucket has completed successfully

Related Pages

• Environment:Mlflow_Mlflow_Python_Runtime_Environment