Managing a fleet of NixOS Part 3 - Welcome to Bento

• Author: Solène
• Date: 04 September 2022
• Tags: bento nixos nix

In this series of articles, I'll explain my steps toward designing an infrastructure to centrally manage a fleet of NixOS systems. The product is now alive and is called Bento.

Comment on Mastodon

Introducing Bento 🥳

I finally wrote an implementation for the NixOS fleet management, it's called Bento.

Bento git project repository

Features

• secure 🛡️: each client can only access its own configuration files (ssh authentication + sftp chroot)
• efficient 🏂🏾: configurations can be built on the central management server to serve binary packages if it is used as a substituters by the clients
• organized 💼: system administrators have all configurations files in one repository to easy management
• peace of mind 🧘🏿: configurations validity can be verified locally by system administrators
• smart 💡: secrets (arbitrary files) can (soon) be deployed without storing them in the nix store
• robustness in mind 🦾: clients just need to connect to a remote ssh, there are many ways to bypass firewalls (corkscrew, VPN, Tor hidden service, I2P, ...)
• extensible 🧰 🪡: you can change every component, if you prefer using GitHub repositories to fetch configuration files instead of a remote sftp server, you can change it
• for all NixOS 💻🏭📱: it can be used for remote workstations, smartphones running NixoS, servers in a datacenter

Evolutions

The project is still bare right now, I started it yesterday and I have many ideas to improve it:

• package it to provide commands in `$PATH` instead of adding scripts to your config repository
• add a rollback features in case an upgrade is losing connectivity
• upgrades can depose a log file in the remote sftp server
• upgrades could be triggered by the user by accessing a local socket, like opening a web page in a web browser to trigger it, if it returns output that'd be better
• provide more useful modules in the utility nix file (automatically use the host as a binary cache for instance)
• have a local information how to ssh to the client to ease the rebuild trigger (like a SSH file containing ssh command line)
• a way to tell a client (when using flakes) to try to update flakes every time even if no configuration changed, to keep them up to date