MLflow Remote Tracking Server

This repository contains a Dockerfile to build a Docker image for the MLflow Remote Tracking Server. The MLflow Remote Tracking Server is a standalone server that can be used to store MLflow experiments and runs in a central location. This allows multiple users to collaborate on the same MLflow experiments and runs.

Their are three main components to the Tracking Server:

• Backend Storage: The backend store is a core component in MLflow Tracking where MLflow stores metadata for Runs and experiments such as: parameters, metrics, tags, and artifacts. We'll be using PostgreSQL for this purpose.
• Artifact Storage: The artifact storage is a shared location where the artifacts for each run are stored. This can be a local directory, an S3 bucket, or an Azure Blob Storage container. We'll be using S3 like storage, MINIO, for this purpose.
• Tracking Server: The tracking server is a RESTful server that provides a central location to log and query MLflow experiments and runs.

The overall architecture of the MLflow Remote Tracking Server is shown below:

Getting Started

Prerequisites

• Docker
• Docker Compose

Running the MLflow Remote Tracking Server

1. Clone this repository:

   git clone https://github.com/AshishSinha5/mlflow-docker.git
   cd mlflow-docker

2. Update the .env file to reflect your desired configuration:

3. Start the MLflow Remote Tracking Server:

   docker-compose up -d

4. In the client code, set the following environment variables to point to the MLflow Remote Tracking Server:

   export MLFLOW_TRACKING_URI= **your machine IP**
   export MLFLOW_S3_ENDPOINT_URL= **your s3 endpoint url**
   export AWS_ACCESS_KEY_ID= **your access key**
   export AWS_SECRET_ACCESS_KEY= **your secret key**

5. You can now access the MLflow UI by navigating to http://your_machine_ip:port in your web browser.