This tool will query a Hydra instance for queue statistics and create or destroy Equinix Metal instances according to load. It is used to scale up and down capacity for hydra.nixos.org.
Currently we have quite a few assumptions that only really apply to the NixOS Hydra instance. Check out the source if you're interested in running this yourself, because you'll probably need to do that anyway ;)
It relies on (this list may not be exhaustive):
-
hydra-packet-importer running to get the instances into Hydra's builder list
-
A Prometheus instance being set up to scrape Hydra, in order to determine what's currently running on a given builder
-
nix-netboot-serve or some other mechanism being set up to serve boot configs for the build machines
