MLflow Tracking Server¶

Custom MLflow Docker image for TradAI platform with PostgreSQL and S3 support.

Features¶

• Based on official ghcr.io/mlflow/mlflow:v2.16.2
• PostgreSQL backend store support (psycopg2 included)
• S3 artifact storage support (boto3 included)
• Non-root user for security
• Health check endpoint
• Production-ready with gunicorn

Quick Start¶

Development (Local Artifacts)¶

docker compose --profile mlflow up

Production (S3 Artifacts)¶

export MLFLOW_BACKEND_STORE_URI="postgresql://user:pass@host:5432/mlflow"
export MLFLOW_DEFAULT_ARTIFACT_ROOT="s3://tradai-mlflow-artifacts/experiments"
docker compose --profile mlflow up

Environment Variables¶

S3 Artifact Configuration¶

For S3 artifacts, ensure AWS credentials are available:

export AWS_ACCESS_KEY_ID=xxx
export AWS_SECRET_ACCESS_KEY=xxx
export AWS_DEFAULT_REGION=eu-central-1
export MLFLOW_DEFAULT_ARTIFACT_ROOT=s3://your-bucket/mlflow

Authentication¶

MLflow supports HTTP Basic Authentication for production deployments.

Option 1: Built-in Auth (MLflow 2.5+)¶

Set environment variables in services/mlflow/.env:

MLFLOW_TRACKING_USERNAME=admin
MLFLOW_TRACKING_PASSWORD=your-secure-password

Option 2: Reverse Proxy Auth¶

For production, use nginx/ALB with authentication in front of MLflow.

Client Configuration¶

Services connect using credentials from .env:

from tradai.common.mlflow_adapter import MLflowAdapter

adapter = MLflowAdapter(
    base_url="https://mlflow.tradai.example.com",
    username=os.getenv("MLFLOW_TRACKING_USERNAME"),
    password=os.getenv("MLFLOW_TRACKING_PASSWORD"),
)

Or set environment variables:

export MLFLOW_TRACKING_URI=https://mlflow.tradai.example.com
export MLFLOW_TRACKING_USERNAME=admin
export MLFLOW_TRACKING_PASSWORD=your-password

Health Check¶

curl http://localhost:5000/health

Build Locally¶

docker build -t tradai/mlflow:latest services/mlflow/

Integration with TradAI Services¶

Services connect to MLflow via MLFLOW_TRACKING_URI:

export MLFLOW_TRACKING_URI=http://mlflow:5000

The MLflowAdapter in tradai-common handles connection and experiment logging.