Skip to content
MLflow Experiment Tracker AI Platform & Data · mlflow 3.17.2 Launch on AWS

Operations

This page covers the MLflow service, consistent backups, moving to another image, and support.

Security model

  • Loopback HTTP. The server listens on 127.0.0.1:5000. The tunnel is how you reach it.
  • Two accounts. ubuntu can read ~/.config/sushi/mlflow/access.json. The administrator identity in /etc/mlflow/owner.json is root-only.
  • Generated secrets. Accounts, the CSRF key, and the databases are created on first boot. The image does not ship a shared password.
  • Artifacts stay on disk. Uploaded artifacts are stored data. The server does not execute them.

Service

sudo systemctl status mlflow sushi-mlflow-firstboot sushi-mlflow-ready
sudo journalctl -u mlflow --since '10 minutes ago'
sudo systemctl restart mlflow

If the first start fails, keep /etc/mlflow, /var/lib/mlflow, and the owner-only files /var/log/mlflow/bootstrap-*.log. Correct the reported error, then restart the first-boot unit and the application unit. Do not delete the databases or regenerate the identity as a repair.

Backups

Stop mlflow.service so the SQLite files are consistent. Back up all of:

  • /var/lib/mlflow
  • /etc/mlflow
  • the customer access file, ~/.config/sushi/mlflow/access.json

Write the archive with mode 0600. Restore the complete set onto the same application version, with the service stopped, and preserve numeric ownership and modes. Do not restore only one of the two databases. Run sushi-mlflow read on a saved manifest before you record new runs.

Updates

A new Marketplace version is a new instance. Restore the full set above onto the same application version, start mlflow, and read back a saved manifest. Keep the archive until that read succeeds.

Support

Email info@softwaresushi.com with the Marketplace version, the Region, and the failing operation. Omit both account files and artifact contents.

The software rate on the listing is USD 0.03 per running instance-hour. EC2, EBS, and data transfer are separate. The MLflow documentation describes the upstream project. Notices for this build are under /opt/mlflow and /usr/share/doc/mlflow.

Last updated October 27, 2026 · Suggest an improvement