MLflow Experiment Tracker

AI Platform & Data mlflow 3.17.2 Hardened AMI

MLflow 3.17.2 on Ubuntu 24.04 records experiment parameters, metrics, and artifacts on this instance. The tracking server and its sign-in database use SQLite, and artifacts stay on local disk. The server runs on CPU. It does not execute uploaded models or pickles.

Get started View on AWS Marketplace

AMI release: 0.1.0 (mlflow 3.17.2)
Recommended: t3.medium
Storage: 32 GiB+ encrypted gp3
Access: Session Manager or SSH
Released: Oct 25, 2026
Delivery: Amazon Machine Image in your own AWS account

Pinned MLflow experiment tracking with local parameters, metrics and artifact storage.

Per-instance application credentials and loopback access through an authenticated tunnel.

A bundled demo and separate readback command verify persisted results.

Quick start

Full guide →

What is on the image¶

This guide matches Marketplace package 0.1.0 (mlflow 3.17.2). Check the listing for the version currently on offer. The listing title is Experiment Tracker for MLflow.

MLflow 3.17.2, the mlflow service, running as the mlflow user on 127.0.0.1:5000.
Two SQLite databases, one for tracking and one for authentication, plus local artifact storage.
A customer account in ~/.config/sushi/mlflow/access.json, readable only by ubuntu, and an administrator identity in /etc/mlflow/owner.json, readable through root.
A reader, sushi-mlflow read, and a walkthrough at /opt/sushi/docs/demo.sh.
Notices under /opt/mlflow and /usr/share/doc/mlflow. Installed inventory is under /opt/sushi/evidence.

Server job execution, periodic job tasks, and custom scorers are disabled. SQLite on one instance is the whole tracking store. It is not a distributed database.

Recommended instance¶

t3.medium (4 GiB of memory) with an encrypted gp3 root volume of at least 32 GiB and IMDSv2. Allow about two minutes after launch for the accounts and the service to initialize. Leave inbound application ports closed.

Billing¶

The listing states a software rate of USD 0.03 per running instance-hour. Confirm the offer before you subscribe. EC2, EBS, and data transfer are billed separately. The supplied demo does not call a model provider.

Tutorial¶

Watch on YouTube.

Where to go next¶

Getting started: launch, record the sample run, and open the UI.
Configuration: account files, what the server will not run, and the tunnel.
Operations: service, backups, and a forgotten administrator password.
Troubleshooting: startup and restore mistakes.

Upstream project¶

MLflow is an open source experiment tracker. Software Sushi packages this image independently and does not claim upstream or AWS endorsement. Delivered notices are under /opt/mlflow and /usr/share/doc/mlflow.

The MLflow documentation describes the upstream project. This image is a single-instance tracking server. It does not serve models, and it does not run server-side jobs or custom scorers.

Last updated October 27, 2026 · Suggest an improvement