MLflow-Docker

A production-ready MLflow Tracking Server running on Docker with MySQL backend and S3-compatible artifact storage (RustFS).

Architecture

┌─────────────────┐     ┌─────────────────┐
│  nginx-proxy    │────▶│     MLflow      │
│   (port 15000)  │     └────────┬────────┘
└─────────────────┘              │
                                 │
                    ┌────────────┴────────────┐
                    │                         │
              ┌───────────┐           ┌───────────────┐
              │   MySQL   │           │    RustFS     │
              │ (Backend) │           │  (Artifacts)  │
              └───────────┘           └───────────────┘

• MLflow Tracking Server - Experiment tracking and model registry
• MySQL - Persistent backend store for metadata
• RustFS - S3-compatible object storage for artifacts
• nginx-proxy - Reverse proxy with Basic Authentication support

Quick Start

1. Configure Environment

cp env.template .env

Edit .env file:

# Hostname(s) for the MLflow server (comma-separated for multiple)
VIRTUAL_HOST=localhost
VIRTUAL_HOST=example.com,localhost

# Optional: Specify MLflow version (leave empty for latest)
MLFLOW_VERSION=

2. Set Up Basic Authentication (Optional)

To enable Basic Authentication, create/edit the htpasswd file:

# Install htpasswd if not available
# macOS: brew install httpd
# Ubuntu: apt-get install apache2-utils

# Create password file (replace 'username' and enter password when prompted)
htpasswd -c ./nginx/htpasswd/localhost username
# If you have multiple hosts, copy the file for each host
cp ./nginx/htpasswd/localhost ./nginx/htpasswd/example.com

3. Start Services

docker compose up -d

The MLflow UI will be available at: http://localhost:15000

4. Stop Services

docker compose down

Usage

Python Client

import mlflow

# Set tracking URI
mlflow.set_tracking_uri("http://localhost:15000")

# If Basic Authentication is enabled
import os
os.environ["MLFLOW_TRACKING_USERNAME"] = "username"
os.environ["MLFLOW_TRACKING_PASSWORD"] = "password"

# Start logging
mlflow.set_experiment("my-experiment")

with mlflow.start_run():
    mlflow.log_param("learning_rate", 0.01)
    mlflow.log_metric("accuracy", 0.95)
    mlflow.log_artifact("model.pkl")

Environment Variables for Remote Access

export MLFLOW_TRACKING_URI="http://your-server:15000"
export MLFLOW_TRACKING_USERNAME="username"
export MLFLOW_TRACKING_PASSWORD="password"

Data Persistence

Data is persisted in the following directories:

• ./mysql/data/ - MySQL database files
• ./rustfs/ - Artifact storage

Backup and Restore database

# Backup
docker compose exec db sh -c 'mysqldump -uroot -p"$MYSQL_ROOT_PASSWORD" "$MYSQL_DATABASE"' > "$(date +%Y%m%d_%H%M%S)_mlflow.sql"
# Restore　(replace the filename with your backup file)
docker compose exec -T db sh -c 'mysql -uroot -p"$MYSQL_ROOT_PASSWORD" "$MYSQL_DATABASE"' < 20251213_143052_mlflow.sql

Migrate S3 bucket

0. Setup access source and destination S3-compatible storage.
1. Run MinIO Client container with host.docker.internal mapping:

docker run --rm -it \
  --add-host=host.docker.internal:host-gateway \
  --entrypoint sh \
  minio/mc

2. Inside the container, configure source and destination:

# Source
mc alias set src http://host.docker.internal:9000 <ACCESS_KEY> <SECRET_KEY>
# Destination
mc alias set dst http://host.docker.internal:9001 <ACCESS_KEY> <SECRET_KEY>

3. Copy the bucket:

mc mirror src/mlflow/artifacts dst/mlflow/artifacts

cleanup removed experiments and runs

docker compose up -d
./mlflow-gc.sh