This directory provides a Docker Compose setup for running MLflow locally with a PostgreSQL backend store and RustFS for S3-compatible artifact storage.

• MLflow Tracking Server
  Serves the REST API and UI (default: http://localhost:5000).

• PostgreSQL
  Stores MLflow's metadata (experiments, runs, params, metrics).

• RustFS Artifact Storage
  Stores model files and run artifacts.

• Git
• Docker and Docker Compose
• On macOS/Windows: Docker Desktop
• On Linux: Docker Engine + compose plugin

Verify installation:

docker --version
docker compose version

git clone https://github.com/mlflow/mlflow.git
cd mlflow/docker-compose/

Copy and customize the environment file:

cp .env.dev.example .env

The .env file defines:

• MLflow server port
• PostgreSQL credentials
• S3 bucket name
• S3-compatible endpoint URL
• Backend-specific configuration for RustFS

Common variables:

• PostgreSQL

  • POSTGRES_USER=mlflow
  • POSTGRES_PASSWORD=mlflow
  • POSTGRES_DB=mlflow

• S3

  • AWS_ACCESS_KEY_ID=s3admin
  • AWS_SECRET_ACCESS_KEY=s3admin
  • AWS_DEFAULT_REGION=us-east-1
  • S3_BUCKET=mlflow

• RustFS

  • RUSTFS_CONSOLE_ENABLE=true

• MLflow

  • MLFLOW_VERSION=latest
  • MLFLOW_HOST=0.0.0.0
  • MLFLOW_PORT=5000
  • MLFLOW_BACKEND_STORE_URI=postgresql://${POSTGRES_USER}:${POSTGRES_PASSWORD}@postgres:5432/${POSTGRES_DB}
  • MLFLOW_ARTIFACTS_DESTINATION=s3://${S3_BUCKET}
  • MLFLOW_S3_ENDPOINT_URL=http://storage:9000

Inside directory mlflow/docker-compose:

docker compose up -d

This will:

• Start PostgreSQL
• Start RustFS
• Start MLflow
• Create the S3 bucket if it doesn't exist

Check status:

docker compose ps

Tail logs:

docker compose logs -f

Once running:

• Open http://localhost:5000 (or the port defined in .env)

You can now log runs, metrics, artifacts, and models to your local MLflow instance.

To stop and remove containers:

docker compose down

To reset everything, including volumes:

docker compose down -v

• Set server domains/host so virtual-hosted requests can be resolved by RustFS:

  RUSTFS_SERVER_DOMAINS=storage:9000

  (match the compose service DNS name)

• Prefer AWS CLI s3api for bucket creation. Some S3 clients default to path-style on custom endpoints; if bucket creation fails with InvalidBucketName, switch to s3api or a client like MinIO mc.

• Inside MLflow, use the internal endpoint:

  MLFLOW_S3_ENDPOINT_URL=http://storage:9000
  MLFLOW_ARTIFACTS_DESTINATION=s3://mlflow/

RustFS usually responds on /health with a json that contains the status of the server:

curl -s http://127.0.0.1:9000/health | grep -q '\"status\"\\s*:\\s*\"ok\"'

Use that in a container healthcheck (no -f, 4xx may appear during bootstrap).

Verify:

• MLFLOW_ARTIFACTS_DESTINATION=s3://<bucket>/
• MLFLOW_S3_ENDPOINT_URL=http://<service>:<port>
• AWS credentials match the backend configuration

To verify S3 storage is working:

aws --endpoint-url=${MLFLOW_S3_ENDPOINT_URL} s3api list-buckets

echo hi > /tmp/t.txt
aws --endpoint-url=${MLFLOW_S3_ENDPOINT_URL} s3 cp /tmp/t.txt s3://${S3_BUCKET}/t.txt
aws --endpoint-url=${MLFLOW_S3_ENDPOINT_URL} s3 cp s3://${S3_BUCKET}/t.txt -

If this passes, MLflow can read and write artifacts to RustFS.

• InvalidBucketName on create-bucket → use s3api (virtual-host friendly) or MinIO mc; ensure RUSTFS_SERVER_DOMAINS matches the S3 hostname.
• Endpoint issues from MLflow → make sure MLFLOW_S3_ENDPOINT_URL uses the service name visible from MLflow (e.g., http://storage:9000).

docker compose down -v
docker compose up -d

docker compose logs -f mlflow
docker compose logs -f postgres
docker compose logs -f storage

Edit .env and restart containers:

docker compose down
docker compose up -d

• Point your training scripts to this server:

  export MLFLOW_TRACKING_URI=http://localhost:5000

• Start logging runs with mlflow.start_run() (Python) or the MLflow CLI.
• Customize the .env and docker-compose.yml to fit your local workflow (e.g., change image tags, add volumes, etc.).

You now have a fully local MLflow stack with persistent metadata and artifact storage—ideal for development and experimentation.