Overview
This AMI runs an experiment tracking server built on the open-source MLflow 3.18.0 project, on Ubuntu 24.04 LTS. Log runs, parameters, metrics and model artifacts from your training code over HTTPS, browse them in the web UI, and let your team share one tracking server instead of comparing spreadsheets. Runs and metadata are stored in PostgreSQL; artifacts (models, files, plots) are stored on the instance and served back through the server. The container image is pinned to version 3.18.0 and baked into the AMI, so first start does not depend on a registry.
What Perimattic adds
Administrator created for you: MLflow has no authentication by default, so every page and API call would otherwise be open to anyone who can reach the server. During first-login setup a random 24-character password is generated on your server and given to the built-in login application before its first start, so the only account is the one you are shown. Setup then checks that the login works, that wrong, default and blank logins are refused, that nobody can self-register, and that anonymous requests to the UI, the API and the artifact store are all refused, before the site is published.
HTTPS and a working end-to-end check: Apache is configured as a reverse proxy and a free Let's Encrypt certificate is requested for your domain with certbot. Setup does not just check that the server answers; it logs a real experiment, run, parameter, metric and artifact with the generated login and reads them back, so you know model tracking actually works before you rely on it.
Pinned and tested: the application version is fixed, not 'latest', so a rebuild never changes the server behind your back. Patched base: Ubuntu 24.04 LTS built through an automated EC2 Image Builder pipeline, with security updates applied at build time.
What you should know
The tracking server used roughly 2.6 GB of memory in our tests (MLflow 3's built-in assistant and job-scheduling features load at start-up), well above what a 4 GB instance can spare alongside PostgreSQL, Apache and the operating system; size accordingly. Artifacts are stored on the instance's own disk, not in object storage, so plan the volume size for your models and files. Usage telemetry is switched off by default and MLflow's telemetry hosts are blocked. Model serving endpoints, the AI assistant and cloud storage back ends are not configured.
Licence and attribution
Built on the open-source MLflow (Apache-2.0 licence), unmodified. Not affiliated with, endorsed by, or sponsored by Databricks or the MLflow project; MLflow is a trademark of its respective owner.
Getting Started Launch the AMI and open ports 22, 80 and 443 in the security group SSH in and answer the setup prompts (domain and admin email) Open https://your-domain/ and sign in with the username and password printed at the end of setup Point your training code at it: export MLFLOW_TRACKING_URI, MLFLOW_TRACKING_USERNAME and MLFLOW_TRACKING_PASSWORD, then log your first run
A domain name with an A record pointing at the instance is required, because the site is served over HTTPS.
This AMI is built and supported by Perimattic.
Book a free setup consultation at https://cal.com/gaurav-pareek-perimattic/marketplace-setup-consultations or email us at aws-support@perimattic.com .
Support Managed and supported by Perimattic, a cloud infrastructure company with 13+ years of experience serving global clients. We offer free setup assistance, custom integrations, infrastructure consulting, and 24/7 support.
Highlights
- MLflow 3.18.0 (Apache-2.0) pinned and pre-installed: experiment tracking with a Postgres backend store and on-instance artifact storage.
- Secure first login: a random 24-character administrator password is generated, sign-up is closed, and every page and API call, including artifact storage, needs the login. HTTPS through Apache and Let's Encrypt.
- Built on patched Ubuntu 24.04 LTS and supported by Perimattic: free setup assistance at aws-support@perimattic.com.
Details
Features and programs
Financing for AWS Marketplace purchases
Pricing
- ...
Dimension | Cost/hour |
|---|---|
t3.medium Recommended | $0.03 |
t3.micro | $0.03 |
c3.2xlarge | $0.03 |
c3.4xlarge | $0.03 |
c3.8xlarge | $0.03 |
c3.large | $0.03 |
c3.xlarge | $0.03 |
c4.2xlarge | $0.03 |
c4.4xlarge | $0.03 |
c4.8xlarge | $0.03 |
Vendor refund policy
For any questions or concerns, please contact: Perimattic.com Email: aws-support@perimattic.com
How can we make this page better?
Legal
Vendor terms and conditions
Content disclaimer
Delivery details
64-bit (x86) Amazon Machine Image (AMI)
Amazon Machine Image (AMI)
An AMI is a virtual image that provides the information required to launch an instance. Amazon EC2 (Elastic Compute Cloud) instances are virtual servers on which you can run your applications and workloads, offering varying combinations of CPU, memory, storage, and networking resources. You can launch as many instances from as many different AMIs as you need.
Version release notes
Initial release.
- Application: Experiment Tracking Server 3.18.0 (container image pinned, baked into the AMI).
- Base: Ubuntu 24.04 LTS, fully patched at build time, built by an automated EC2 Image Builder pipeline.
- TLS: Apache reverse proxy with a Let's Encrypt certificate requested by certbot at first login.
- Credentials are generated on the instance at first login; none are present in the image.
- The administrator is created in MLflow's own built-in login application with a random 24-character password generated at first login; sign-up is closed, and setup verifies the login, refusal of wrong/default/blank/anonymous access, and a real run/metric/artifact round trip before the site is published.
- MLflow 3.18.0 (Apache-2.0, vendor image plus the PostgreSQL driver) runs as an unprivileged user on the loopback interface; usage telemetry is switched off and MLflow's telemetry hosts are blocked.
- The tracking server used about 2.6 GB of memory in testing; a t3.large or larger instance is recommended.
Additional details
Usage instructions
-
Launch an Instance
From AWS Marketplace, click Continue to Subscribe and then Continue to Configuration to launch the AMI.
-
Choose the Recommended Instance Type
t3.large (8 GB of memory) is the recommended size: the tracking server alone used about 2.6 GB in our tests, which a 4 GB instance cannot spare alongside PostgreSQL, Apache and the operating system.
-
Configure Security Group
Allow inbound traffic on the following ports:
- 22 (SSH) - terminal access for setup
- 80 (HTTP) - needed to validate your domain for the TLS certificate
- 443 (HTTPS) - access to Experiment Tracking Server
-
Point a domain at the instance
Create a DNS A record (for example tracking.example.com) pointing to the instance's public IP before you log in.
-
Connect to the Instance
ssh -i <your-key.pem> ubuntu@<public-ip>
-
First-Login Setup
The first SSH login runs a guided setup:
- Prompts for your domain and admin email and checks the DNS A record
- Generates a random 24-character administrator password on your server
- Starts PostgreSQL and the tracking server with the built-in login application configured and that password set, then checks the login works, that wrong, default and blank logins and anonymous requests are refused, and that a run, metric and artifact can be logged and read back
- Configures Apache as an HTTPS reverse proxy and requests a Let's Encrypt certificate
- Prints the URL, username and password; they are also stored in /opt/app/access.txt (root only)
If a step fails, fix the problem shown and log in again: the setup resumes where it stopped.
-
Access Experiment Tracking Server
Open https://<your-domain>/ and sign in with the username and password printed at the end of setup (view them again with: sudo cat /opt/app/access.txt). Set MLFLOW_TRACKING_URI, MLFLOW_TRACKING_USERNAME and MLFLOW_TRACKING_PASSWORD in your training environment, then log runs as usual with the mlflow client library.
-
Credentials
Generated on your server, never shared between customers. View them again with:
sudo cat /opt/app/access.txt
-
Manage the Services
cd /opt/app sudo docker compose ps sudo docker compose restart
Artifacts (models, files, plots) are in /opt/app/artifacts and run/metric data is in PostgreSQL under /opt/app/postgres; both are on the root volume with no size limit, so size the volume for your models and watch disk usage. Back up or snapshot /opt/app (stop the stack first for a consistent copy).
Thank you for subscribing.
Need Support or Customization? Perimattic's team can help with setup issues, custom features and integrations:
Email us: aws-support@perimattic.com WhatsApp: +91-92142 66896 Learn more: <www.perimattic.com >
We offer free setup assistance, custom development and integrations, and infrastructure consulting.
Resources
Vendor resources
Support
Vendor support
Support by Perimattic
Perimattic provides 24/7 support for all AWS Marketplace products. Our team has 13+ years of cloud infrastructure expertise serving global clients.
Contact Channels: Email: aws-support@perimattic.com WhatsApp: +91-92142 66896 Website: <www.perimattic.com >
What We Help With: Initial setup and configuration assistance (free) Troubleshooting connectivity, performance, and backup issues Custom development and integrations Infrastructure consulting and architecture review Migration planning from other solutions Automation solutions and scaling guidance
Instance Sizing Guidance: t2.medium or t3.medium: Development environments, small applications m5.large: Production workloads with moderate traffic m5.xlarge: High-traffic production deployments and larger datasets
For workload-specific sizing recommendations, contact our team for a free consultation.
Refunds and Issues: If you experience any issues or need to request a refund, contact us at aws-support@perimattic.com and we will respond promptly.
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.