Deploy and Host MLflow Full on Railway

MLflow Full (mlflow:v3.10.1-full) is a platform for managing the machine learning and GenAI lifecycle. It provides experiment tracking, model registry, artifact storage, and prompt observability in a unified interface. Teams can log parameters, metrics, prompts, and models while organizing experiments and managing versions of ML and LLM systems.

When deployed on Railway, MLflow runs as a tracking server accessible through a web interface. Experiment metadata is stored in a PostgreSQL database, while artifacts (models, logs, datasets, evaluation outputs) should be stored in an S3-compatible bucket for persistence.

MLflow Artifact Storage

Currently, artifact storage is not persistent.

Artifacts are saved locally inside the container at:

/app/mlruns

Because Railway containers use ephemeral storage, artifacts may be lost after redeploys or restarts.

To enable persistent artifact storage, configure an S3-compatible bucket (AWS S3, Cloudflare R2, MinIO, etc.) using the following environment variables:

BACKEND_S3=s3://mlflow-artifacts/

AWS_ACCESS_KEY_ID=
AWS_SECRET_ACCESS_KEY=
AWS_DEFAULT_REGION=

The MLflow server is started with:

--default-artifact-root ${BACKEND_S3:-$BACKEND_s3}

If BACKEND_S3 is defined, MLflow automatically detects and uses it as the artifact bucket.

Dependencies

PostgreSQL database for experiment metadata
S3-compatible storage bucket for artifacts

Deployment References

MLflow Documentation
https://mlflow.org/docs/latest

Railway Deployment Platform
https://railway.app

Oploy AI & Data Science Platform
https://www.oploy.eu