Kyle Sferrazza

notes, posts, /etc

Using Nixops with Google Compute Engine

Keeping Terraform and NixOps in sync, while still allowing regular SSH access to the instance.

3 minute read

The Base Image

Cloud Compute does not have a predefined image for NixOS, which is the distribution we’re going to install on the instance.

I’ve created a simple NixOS system configuration and terraform script (see the code on GitHub here) that builds that system configuration, uploads it to cloud compute, and sets up an f1-micro instance using the image. f1-micro is free for 1 instance per month on the Google Cloud free tier.

The config for the system itself is quite small:

  imports = [ <nixpkgs/nixos/modules/virtualisation/google-compute-image.nix> ];
  users.users.root.openssh.authorizedKeys.keyFiles = [ ~/.ssh/ ];

This config imports the base Cloud Compute image from nixpkgs, which sets up some sane defaults. It then makes sure we are able to log in as root after the instance is created, by copying over the current user’s public key and adding it to root’s list of authorized SSH keys.

Deploying the Initial Image with Terraform

So, to generate the image and provision a Compute instance, clone the repo and run terraform apply. The terraform script is configured to take inputs for various things that may change from installation to installation, like the user name and location of the credentials file (learn how to generate the credentials file in the terraform docs).

Terraform is smart and will automatically check the latest state of all of the resources you have defined and get them from their current state to the desired state when you run terraform apply. You can also use terraform plan to see what changes terraform will make before it makes them.

Deploying a Project with NixOps

Now that NixOS is successfully running on a Cloud Compute instance, we can get started provisioning it with NixOps. NixOps lets you use the same ideas and configurations you use for configuring a NixOS system to deploy applications and services to remote systems, virtual machines, and cloud providers.

The NixOps intro is worth reading if you haven’t read it already and plan to use NixOps to deploy your applications. It goes over the benefits and features of NixOps, and gives a high-level overview of why you might want to use it.

From the project you wish to deploy (I have a sample project for hosting containers), define a NixOS system config that runs your application. I recommend putting this config in one file (say system.nix), and making another file called ops.nix that will be loaded into NixOps. The structure of the ops.nix file is an attribute set where the keys are hostnames/IPs and the values are NixOS system configurations.

Take the ops.nix from my container-host project, for example:

  "" = import ./server.nix;

This file just says to NixOps: “build the system configuration in server.nix and deploy it to the machine at”.

Now, go ahead and create the NixOps deployment by running nixops create -d my-deployment-name ops.nix. Once the deployment is created, deploy (or redeploy) with nixops deploy -d my-deployment-name. NixOps will replace root SSH keys by default, which is usually okay since you can still use nixops ssh -d my-deployment-name hostname to SSH into the machine, but I make sure to keep the trick from above in server.nix so I can still use SSH as usual (or set up multiple NixOps deployments to the same machine):

users.users.root.openssh.authorizedKeys.keyFiles = [ ~/.ssh/ ];

If all went well, you should hopefully be all set up using NixOS and NixOps on the free tier of Google Cloud Compute!