Paranoid NixOS on AWS

Published on , 4622 words, 17 minutes to read

In the last post we covered a lot of the base groundwork involved in making a paranoid NixOS setup. Today we're gonna throw this into prod by making a base NixOS image with it.

Cadey is coffee
<Cadey>

Normally I don't suggest people throw these things into production directly, if only to have some kind of barrier between you and your money generator; however today is different. It's probably not completely unsafe to put this in production, but I really would suggest reading and understanding this article before doing so.

At a high level we are going to do the following:

What is an Image?

Before we yolo this all into prod, let's cover what we're actually doing. There are a lot of conflicting buzzwords here, so I'm going to go out of my way to attempt to simplify them down so that we use my arbitrary definitions of buzzwords instead of what other people will imply they mean. You're reading my blog, you get my buzzwords; it's as simple as that.

In this post we are going to create a base system that you can build your production systems on top of. This base system will be crystallized into an image that AWS will use as the initial starting place for servers.

Mara is hmm
<Mara>

So you create the system definition for your base system, then turn that into an image and put that image into AWS?

Cadey is enby
<Cadey>

Yep! The exact steps are a little more complicated but at a high level that's what we're doing.

Base Setup

I'm going to be publishing my work for this post here, but you can follow along in this post to understand the individual steps here.

First, let's set up the environment with lorri and niv. Lorri will handle creating a cached nix-shell environment for us to run things in and niv will handle pinning NixOS to an exact version so you can get a more reproducible production environment.

Set up lorri:

$ lorri init
Aug 11 09:41:50.966 INFO wrote file, path: ./shell.nix
Aug 11 09:41:50.966 INFO wrote file, path: ./.envrc
Aug 11 09:41:50.966 INFO done
direnv: error /home/cadey/code/cadey/paranix-configs/.envrc is blocked. Run `direnv allow` to approve its content
$ direnv allow
direnv: loading ~/code/cadey/paranix-configs/.envrc
Aug 11 09:41:54.581 INFO lorri has not completed an evaluation for this project yet, nix_file: /home/cadey/code/cadey/paranix-configs/shell.nix
direnv: export +IN_NIX_SHELL
Mara is hacker
<Mara>

Why are you putting the $ before every command in these examples? It looks extraneous to me.

Cadey is enby
<Cadey>

The $ is there for two main reasons. First, it allows there to be a clear delineation between the commands being typed and their output. Secondly it makes it slightly harder to blindly copy this into your shell without either editing the $ out or selecting around it. My hope is that this will make you read the command and carefully consider whether or not you actually want to run it.

Set up niv:

$ niv init
Initializing
  Creating nix/sources.nix
  Creating nix/sources.json
  Importing 'niv' ...
  Adding package niv
    Writing new sources file
  Done: Adding package niv
  Importing 'nixpkgs' ...
  Adding package nixpkgs
    Writing new sources file
  Done: Adding package nixpkgs
Done: Initializing
Mara is hacker
<Mara>

If you don't already have niv in your environment, you can hack around that by running all the niv commands before you set up shell.nix like this:

nix-shell -p niv --run 'niv blah'

And finally pin nixpkgs to a specific version of NixOS.

Mara is hacker
<Mara>

At the time of writing this article, NixOS 21.05 is the stable release, so that is what is used here. Check what the newest version is for you.

$ niv update nixpkgs -b nixos-21.05
Update nixpkgs
Done: Update nixpkgs
$

This will become the foundation of our NixOS systems and production images.

You should then set up your shell.nix to look like this:

let
  sources = import ./nix/sources.nix;
  pkgs = import sources.nixpkgs { };
in pkgs.mkShell {
  buildInputs = with pkgs; [
    niv
    terraform

    bashInteractive
  ];
};

Set Up Unix Accounts

Mara is hacker
<Mara>

This step can be omitted if you are grafting this into an existing NixOS configs repository, however it would be good to read through this to understand the directory layout at play here.

It's probably important to be able to have access to production machines. Let's create a NixOS module that will allow you to SSH into the machine. In your paranix-configs folder, run this command to make a common config directory:

$ mkdir common
$ cd common

Now in that common directory, open default.nix in emacs your favorite text editor and copy in this skeleton:

# common/default.nix

{ config, lib, pkgs, ... }:

{
  imports = [ ./users.nix ];

  nix.autoOptimiseStore = true;

  users.users.root.openssh.authorizedKeys.keys = [ "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIPg9gYKVglnO2HQodSJt4z4mNrUSUiyJQ7b+J798bwD9" ];

  services.tailscale.enable = true;

  # Tell the firewall to implicitly trust packets routed over Tailscale:
  networking.firewall.trustedInterfaces = [ "tailscale0" ];

  security.auditd.enable = true;
  security.audit.enable = true;
  security.audit.rules = [
    "-a exit,always -F arch=b64 -S execve"
  ];

  security.sudo.execWheelOnly = true;
  environment.defaultPackages = lib.mkForce [];

  services.openssh = {
    passwordAuthentication = false;
    allowSFTP = false; # Don't set this if you need sftp
    challengeResponseAuthentication = false;
    extraConfig = ''
      AllowTcpForwarding yes
      X11Forwarding no
      AllowAgentForwarding no
      AllowStreamLocalForwarding no
      AuthenticationMethods publickey
    '';
  };

  # PCI compliance
  environment.systemPackages = with pkgs; [ clamav ];
}
Mara is hacker
<Mara>

Astute readers will notice that this is less paranoid than the last post. This was pared down after private feedback.

This will create common as a folder that can be imported as a NixOS module with some basic settings and then tells NixOS to try importing users.nix as a module. This module doesn't exist yet, so it will fail when we try to import it. Let's fix that by making users.nix:

# common/users.nix

{ config, lib, pkgs, ... }:

with lib;

let
  # These options will be used for user account defaults in
  # the `mkUser` function.
  xeserv.users = {
    groups = mkOption {
      type = types.listOf types.str;
      default = [ "wheel" ];
      example = ''[ "wheel" "libvirtd" "docker" ]'';
      description =
        "The Unix groups that Xeserv staff users should be assigned to";
    };

    shell = mkOption {
      type = types.package;
      default = pkgs.bashInteractive;
      example = "pkgs.powershell";
      description =
        "The default shell that Xeserv staff users will be given by default.";
    };
  };

  cfg = config.xeserv.users;

  mkUser = { keys, shell ? cfg.shell, extraGroups ? cfg.groups, ... }: {
    isNormalUser = true;
    inherit extraGroups shell;
    openssh.authorizedKeys = {
      inherit keys;
    };
  };
in {
  options.xeserv.users = xeserv.users;

  config.users.users = {
    cadey = mkUser {
      keys = [ "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIPg9gYKVglnO2HQodSJt4z4mNrUSUiyJQ7b+J798bwD9" ];
    };
    twi = mkUser {
      keys = [ "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIPYr9hiLtDHgd6lZDgQMkJzvYeAXmePOrgFaWHAjJvNU" ];
    };
  };
}
Mara is hacker
<Mara>

It's worth noting that xeserv in there can be anything you want. It's set to xeserv as we are imagining that this is for the production environment of a company named Xeserv.

Paranoid Settings

Next we're going to set up the paranoid settings from the last post into a module named paranoid.nix. First we'll need to grab impermanence into our niv manifest like this:

$ niv add nix-community/impermanence
Adding package impermanence
  Writing new sources file
Done: Adding package impermanence

Then open common/default.nix and change this line:

imports = [ ./users.nix ];

To something like this:

imports = [ ./paranoid.nix ./users.nix ];

Then open ./paranoid.nix in a text editor and paste in the following:

# common/paranoid.nix

{ config, pkgs, lib, ... }:

with lib;

let
  sources = import ../nix/sources.nix;
  impermanence = sources.impermanence;
  cfg = config.xeserv.paranoid;

  ifNoexec = if cfg.noexec then [ "noexec" ] else [ ];
in {
  imports = [ "${impermanence}/nixos.nix" ];

  options.xeserv.paranoid = {
    enable = mkEnableOption "enables ephemeral filesystems and limited persistence";
    noexec = mkEnableOption "enables every mount on the system save /nix being marked as noexec (potentially dangerous at a social level)";
  };

  config = mkIf cfg.enable {
    fileSystems."/" = mkForce {
      device = "none";
      fsType = "tmpfs";
      options = [ "defaults" "size=2G" "mode=755" ] ++ ifNoexec;
    };

    fileSystems."/etc/nixos".options = ifNoexec;
    fileSystems."/srv".options = ifNoexec;
    fileSystems."/var/lib".options = ifNoexec;
    fileSystems."/var/log".options = ifNoexec;

    fileSystems."/boot" = {
      device = "/dev/disk/by-label/boot";
      fsType = "vfat";
    };

    fileSystems."/nix" = {
      device = "/dev/disk/by-label/nix";
      autoResize = true;
      fsType = "ext4";
    };

    boot.cleanTmpDir = true;

    environment.persistence."/nix/persist" = {
      directories = [
        "/etc/nixos" # nixos system config files, can be considered optional
        "/srv" # service data
        "/var/lib" # system service persistent data
        "/var/log" # the place that journald dumps it logs to
      ];
    };

    environment.etc."ssh/ssh_host_rsa_key".source =
      "/nix/persist/etc/ssh/ssh_host_rsa_key";
    environment.etc."ssh/ssh_host_rsa_key.pub".source =
      "/nix/persist/etc/ssh/ssh_host_rsa_key.pub";
    environment.etc."ssh/ssh_host_ed25519_key".source =
      "/nix/persist/etc/ssh/ssh_host_ed25519_key";
    environment.etc."ssh/ssh_host_ed25519_key.pub".source =
      "/nix/persist/etc/ssh/ssh_host_ed25519_key.pub";
    environment.etc."machine-id".source = "/nix/persist/etc/machine-id";
  };
}

This should give us the base that we need to build the system image for AWS.

Building The Image

As I mentioned earlier we need to build a system image before we can build the image. NixOS normally hides a lot of this magic from you, but we're going to scrape away all that magic and do this by hand. In your paranix-configs folder, create a folder named images. This creatively named folder is where we will store our NixOS image generation scripts.

Copy this code into build.nix. This will tell NixOS to create a new system closure with configuration in images/configuration.nix:

# images/build.nix

let
  sources = import ../nix/sources.nix;
  pkgs = import sources.nixpkgs { };
  sys = (import "${sources.nixpkgs}/nixos/lib/eval-config.nix" {
    system = "x86_64-linux";
    modules = [ ./configuration.nix ];
  });
in sys.config.system.build.toplevel

And in images/configuration.nix add this skeleton config:

# images/configuration.nix

{ config, pkgs, lib, modulesPath, ... }:

{
  imports = [ ../common (modulesPath + "/virtualisation/amazon-image.nix") ];

  xeserv.paranoid.enable = true;
}
Mara is hacker
<Mara>

You can adapt this to other clouds by changing what module is imported. See the list of available modules here.

Then you can kick off the build with nix-build:

$ nix-build build.nix

It will take a moment to assemble everything together and when you are done you should have an entire functional system closure in ./result:

$ cat ./result/nixos-version
21.05pre-git
Mara is hacker
<Mara>

It has pre-git here because we're using a pinned commit of the nixos-21.05 git branch. Release channels don't have that suffix there.

From here we need to put this base system closure into a disk image for AWS. This process is a bit more involved, but here are the high level things needed to make a disk image for NixOS (or any Linux system for that matter):

We can model this using a Nix function. This function would need to take in the system config, some metadata about the kind of image to make and then it would build the image and return the result. I've made this available here so you can grab it into your config folder like this:

$ wget -O make-image.nix https://tulpa.dev/cadey/paranix-configs/raw/branch/main/images/make-image.nix

Then we can edit build.nix to look like this:

# images/build.nix

let
  sources = import ../nix/sources.nix;
  pkgs = import sources.nixpkgs { };
  config = (import "${sources.nixpkgs}/nixos/lib/eval-config.nix" {
    system = "x86_64-linux";
    modules = [ ./configuration.nix ];
  });

in import ./make-image.nix {
  inherit (config) config pkgs;
  inherit (config.pkgs) lib;
  format = "vpc"; # change this for other clouds
}

Then you can build the AWS image with nix-build:

$ nix-build build.nix

This will emit the AWS disk image in ./result:

$ ls ./result/
nixos.vhd
Mara is hacker
<Mara>

AWS uses Microsoft Virtual PC hard disk files as the preferred input for their vmimport service. This is probably a legacy thing.

Terraforming

Terraform is not my favorite tool on the planet, however it is quite useful for beating AWS and other clouds into shape. We will be using Terraform to do the following:

This sounds like a lot, but it's really not as much as it sounds. A lot of this is boilerplate. The cost associated with these steps should be minimal.

In the root of your paranix-configs folder, make a folder called terraform, as this is where our terraform configuration will live:

$ mkdir terraform
$ cd terraform

Then you can proceed to the following steps.

S3 State Bucket

In that folder, make a folder called bootstrap, this configuration will contain the base S3 bucket config for Terraform state:

$ mkdir bootstrap
$ cd bootstrap

Copy this terraform code into main.tf:

# terraform/bootstrap/main.tf

provider "aws" {
  region = "us-east-1"
}

resource "aws_s3_bucket" "bucket" {
  bucket = "xeserv-tf-state-paranix"
  acl    = "private"

  tags = {
    Name = "Terraform State"
  }
}

Then run terraform init to set up the terraform environment:

$ terraform init

It will download the AWS provider and run a few tests on your config to make sure things are correct. Once this is done, you can run terraform plan:

$ terraform plan
Terraform used the selected providers to generate the following execution plan. Resource actions
are indicated with the following symbols:
  + create

Terraform will perform the following actions:

  # aws_s3_bucket.bucket will be created
  + resource "aws_s3_bucket" "bucket" {
      + acceleration_status         = (known after apply)
      + acl                         = "private"
      + arn                         = (known after apply)
      + bucket                      = "xeserv-tf-state-paranoid"
      + bucket_domain_name          = (known after apply)
      + bucket_regional_domain_name = (known after apply)
      + force_destroy               = false
      + hosted_zone_id              = (known after apply)
      + id                          = (known after apply)
      + region                      = (known after apply)
      + request_payer               = (known after apply)
      + tags                        = {
          + "Name" = "Terraform State"
        }
      + tags_all                    = {
          + "Name" = "Terraform State"
        }
      + website_domain              = (known after apply)
      + website_endpoint            = (known after apply)

      + versioning {
          + enabled    = (known after apply)
          + mfa_delete = (known after apply)
        }
    }

Plan: 1 to add, 0 to change, 0 to destroy.

Note: You didn't use the -out option to save this plan, so Terraform can't guarantee to take
exactly these actions if you run "terraform apply" now.

Terraform is very pedantic about what the state of the world is. In this case nothing in the associated state already exists, so it is saying that it needs to create the S3 bucket that we will use for our Terraform states in the future. We can apply this with terraform apply:

$ terraform apply
<the same thing as the plan>

Do you want to perform these actions?
  Terraform will perform the actions described above.
  Only 'yes' will be accepted to approve.

  Enter a value:

If you want to perform these actions, follow the instructions.

  Enter a value: yes

aws_s3_bucket.bucket: Creating...
aws_s3_bucket.bucket: Creation complete after 3s [id=xeserv-tf-state-paranoid]

Apply complete! Resources: 1 added, 0 changed, 0 destroyed.

Now that we have the state bucket, let's use it to create our AMI.

Creating the AMI

In your terraform folder, create a new folder called aws_image. This is where the terraform configuration for uploading our disk image to AWS will live.

$ mkdir aws_image
$ cd aws_image
Mara is hacker
<Mara>

This part of the config is modified from the instructions on how to create an AMI from a locally created VM image

here

.

Make a file called main.tf and we'll add to it as we go through this section.

In main.tf, add the following boilerplate to make the AWS provider use the terraform state bucket we just created:

# terraform/aws_image/main.tf

provider "aws" {
  region = "us-east-1"
}

terraform {
  backend "s3" {
    bucket = "xeserv-tf-state-paranoid"
    key    = "aws_image"
    region = "us-east-1"
  }
}

This will tell the AWS provider to use the S3 bucket we just made, but also to put the terraform state in a key called aws_image. We will reuse this state later for making our printer facts host. After we do this, we should run terraform init to make sure that the state bucket is working:

$ terraform init
Initializing the backend...

Initializing provider plugins...
- Finding latest version of hashicorp/aws...
- Installing hashicorp/aws v3.53.0...
- Installed hashicorp/aws v3.53.0 (signed by HashiCorp)

Terraform has created a lock file .terraform.lock.hcl to record the provider
selections it made above. Include this file in your version control repository
so that Terraform can guarantee to make the same selections by default when
you run "terraform init" in the future.

Terraform has been successfully initialized!

You may now begin working with Terraform. Try running "terraform plan" to see
any changes that are required for your infrastructure. All Terraform commands
should now work.

If you ever set or change modules or backend configuration for Terraform,
rerun this command to reinitialize your working directory. If you forget, other
commands will detect it and remind you to do so if necessary.

Now let's create the S3 bucket that we will put our NixOS image in:

# terraform/aws_image/main.tf

resource "aws_s3_bucket" "images" {
  bucket = "xeserv-ami-images"
  acl    = "private"

  tags = {
    Name = "Xeserv AMI Images"
  }
}

Then let's create the IAM role and policy that allows the VM importer service to import objects from S3 into EBS snapshots that we use to create an AMI.

In the aws_image folder, copy this trust policy statement into vmie-trust-policy.json:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Principal": { "Service": "vmie.amazonaws.com" },
      "Action": "sts:AssumeRole",
      "Condition": {
        "StringEquals": {
          "sts:Externalid": "vmimport"
        }
      }
    }
  ]
}

This will be used to give the VM import service permission to act against AWS on your behalf.

In main.tf, add the following role and policy to the configuration:

# terraform/aws_image/main.tf

resource "aws_iam_role" "vmimport" {
  name               = "vmimport"
  assume_role_policy = file("./vmie-trust-policy.json")
}

resource "aws_iam_role_policy" "vmimport_policy" {
  name   = "vmimport"
  role   = aws_iam_role.vmimport.id
  policy = <<EOF
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "s3:ListBucket",
        "s3:GetObject",
        "s3:GetBucketLocation"
      ],
      "Resource": [
        "${aws_s3_bucket.images.arn}",
        "${aws_s3_bucket.images.arn}/*"
      ]
    },
    {
      "Effect": "Allow",
      "Action": [
        "s3:GetBucketLocation",
        "s3:GetObject",
        "s3:ListBucket",
        "s3:PutObject",
        "s3:GetBucketAcl"
      ],
      "Resource": [
        "${aws_s3_bucket.images.arn}",
        "${aws_s3_bucket.images.arn}/*"
      ]
    },
    {
      "Effect": "Allow",
      "Action": [
        "ec2:ModifySnapshotAttribute",
        "ec2:CopySnapshot",
        "ec2:RegisterImage",
        "ec2:Describe*"
      ],
      "Resource": "*"
    }
  ]
    }
EOF
}
Mara is hmm
<Mara>

Why do you define the trust policy in an external file but you have the role policy defined inline?

Cadey is enby
<Cadey>

Look at the Resources defined in the Statement list. The S3 bucket in question needs to be defined explicitly by its

ARN

, and in order to give the vmimport service the minimal possible permissions, we need to template out that policy JSON file, and doing this inline in Terraform is a lot simpler.

And now we should run terraform plan and terraform apply to make sure everything works okay:

$ terraform plan
<omitted>
Plan: 3 to add, 0 to change, 0 to destroy.

$ terraform apply
<omitted>
Apply complete! Resources: 3 added, 0 changed, 0 destroyed.

Perfect! Now we need to upload the image to S3. You are going to have to build the NixOS image outside of terraform, so run nix-build:

$ nix-build ../../build.nix

This should largely be a no-op and will put the correct result symlink in your aws_image folder so terraform can read the image metadata.

Mara is hacker
<Mara>

Practically you would want to make a script to run terraform, and in the script for this folder you would probably want to add that nix-build command to that script. However this is trivial and is thus an exercise for the reader.

In your main.tf file, add this:

# terraform/aws_image/main.tf

resource "aws_s3_bucket_object" "nixos_21_05" {
  bucket = aws_s3_bucket.images.bucket
  key    = "nixos-21.05-paranoid.vhd"

  source = "./result/nixos.vhd"
  etag   = filemd5("./result/nixos.vhd")
}

Now we need to create the EBS snapshot. Copy this into your main.tf:

# terraform/aws_image/main.tf

resource "aws_ebs_snapshot_import" "nixos_21_05" {
  disk_container {
    format = "VHD"
    user_bucket {
      s3_bucket = aws_s3_bucket.images.bucket
      s3_key    = aws_s3_bucket_object.nixos_21_05.key
    }
  }

  role_name = aws_iam_role.vmimport.name

  tags = {
    Name = "NixOS-21.05"
  }
}

This step may take a while (more than 5 minutes), so let's run terraform plan and then terraform apply:

$ terraform plan
Plan: 2 to add, 0 to change, 0 to destroy.

$ terraform apply
Apply complete! Resources: 2 added, 0 changed, 0 destroyed.

Finally you can create the AMI and export the AMI ID like this:

# terraform/aws_image/main.tf

resource "aws_ami" "nixos_21_05" {
  name                = "nixos_21_05"
  architecture        = "x86_64"
  virtualization_type = "hvm"
  root_device_name    = "/dev/xvda"
  ena_support         = true
  sriov_net_support   = "simple"

  ebs_block_device {
    device_name           = "/dev/xvda"
    snapshot_id           = aws_ebs_snapshot_import.nixos_21_05.id
    volume_size           = 40 # you can go as low as 8 GB, but 40 is a nice number
    delete_on_termination = true
    volume_type           = "gp3"
  }
}

output "nixos_21_05_ami" {
  value = aws_ami.nixos_21_05.id
}

Then run terraform plan and terraform apply:

$ terraform plan
Plan: 1 to add, 0 to change, 0 to destroy.

$ terraform apply
Apply complete! Resources: 1 added, 0 changed, 0 destroyed.

Outputs:

nixos_21_05_ami = "ami-0f43f74cbbdd1ddef"

Et voila! We have a NixOS base image that we can use for production workloads. Let's use it to create a NixOS server running the printer facts service.

Mara is hacker
<Mara>

KEEP IN MIND that this configuration means that every time you rebuild and upload this image you potentially risk breaking production machines. Don't rebuild this config more than once every 6 months (or when you bump to a new release of NixOS) at most.

Using the AMI

Let's make a new folder in the terraform folder called printerfacts. In this folder we're going to set up a new terraform state that imports the AMI state we just made and then we will use that AMI to run the printer facts service.

$ mkdir printerfacts
$ cd printerfacts

In main.tf, copy the following:

# terraform/printerfacts/main.tf

provider "aws" {
  region = "us-east-1"
}

terraform {
  backend "s3" {
    bucket = "xeserv-tf-state-paranoid"
    key    = "printerfacts"
    region = "us-east-1"
  }
}

Now you can terraform init as normal to ensure everything is working as we expect:

$ terraform init
Terraform has been successfully initialized!

Then let's add the aws_image state as a data source. This will let us reference the AMI ID from the remote state file instead of having to build it from scratch every time.

# terraform/printerfacts/main.tf

data "terraform_remote_state" "aws_image" {
  backend = "s3"

  config = {
    bucket = "xeserv-tf-state-paranoid"
    key    = "aws_image"
    region = "us-east-1"
  }
}

AWS wants us to create a keypair for the instance, so to make AWS happy we will make a keypair like this:

# terraform/printerfacts/main.tf

resource "tls_private_key" "state_ssh_key" {
  algorithm = "RSA"
}

resource "aws_key_pair" "generated_key" {
  key_name   = "generated-key-${sha256(tls_private_key.state_ssh_key.public_key_openssh)}"
  public_key = tls_private_key.state_ssh_key.public_key_openssh
}
Mara is hacker
<Mara>

You will need to run terraform init after this step.

Now we need to create a security group for this instance. This security group should do the following:

You can do this with this terraform fragment:

# terraform/printerfacts/main.tf

resource "aws_security_group" "printerfacts" {
  ingress {
    from_port   = 22
    to_port     = 22
    protocol    = "tcp"
    cidr_blocks = ["0.0.0.0/0"]
  }
  ingress {
    from_port   = 80
    to_port     = 80
    protocol    = "tcp"
    cidr_blocks = ["0.0.0.0/0"]
  }
  ingress {
    from_port   = -1
    to_port     = -1
    protocol    = "icmp"
    cidr_blocks = ["0.0.0.0/0"]
  }
  egress {
    from_port   = -1
    to_port     = -1
    protocol    = "icmp"
    cidr_blocks = ["0.0.0.0/0"]
  }
  egress {
    from_port   = 0
    to_port     = 65535
    protocol    = "tcp"
    cidr_blocks = ["0.0.0.0/0"]
  }
  egress {
    from_port   = 0
    to_port     = 65535
    protocol    = "udp"
    cidr_blocks = ["0.0.0.0/0"]
  }
}

Then we can create the AWS instance using our AMI, keypair and security group:

# terraform/printerfacts/main.tf

resource "aws_instance" "printerfacts" {
  ami           = data.terraform_remote_state.aws_image.outputs.nixos_21_05_ami
  instance_type = "t3.micro"
  security_groups = [
    aws_security_group.printerfacts.name,
  ]
  key_name = aws_key_pair.generated_key.key_name

  root_block_device {
    volume_size = 40 # GiB
  }

  tags = {
    Name = "xe-printerfacts"
  }
}

And then we can create a NixOS deploy config with the fantastic deploy_nixos module from Tweag. Copy this into your main.tf:

# terraform/printerfacts/main.tf

module "deploy_printerfacts" {
  source          = "git::https://github.com/Xe/terraform-nixos.git//deploy_nixos?ref=1b49f2c6b4e7537cca6dd6d7b530037ea81e8268"
  nixos_config    = "${path.module}/printerfacts.nix"
  hermetic        = true
  target_user     = "root"
  target_host     = aws_instance.printerfacts.public_ip
  ssh_private_key = tls_private_key.state_ssh_key.private_key_pem
  ssh_agent       = false
  build_on_target = false
}
Mara is hacker
<Mara>

You will need to run terraform init again after this step.

Now let's make the printerfacts.nix host definition. We're going to start with a simple config to begin with. This will start nginx in a mostly broken but still semi-functional state on port 80.

# terraform/printerfacts/printerfacts.nix

let
  sources = import ../../nix/sources.nix;
  pkgs = import sources.nixpkgs { };
  system = "x86_64-linux";

  configuration = { config, lib, pkgs, ... }: {
    imports = [
      ../../common
      "${sources.nixpkgs}/nixos/modules/virtualisation/amazon-image.nix"
    ];

    networking.firewall.allowedTCPPorts = [ 22 80 ];

    xeserv.paranoid.enable = true;

    services.nginx.enable = true;
  };
in import "${sources.nixpkgs}/nixos" { inherit system configuration; }
Mara is hmm
<Mara>

What is up with that config? It doesn't look like a normal NixOS module at all.

Cadey is enby
<Cadey>

That is a NixOS config that will use the pinned version of nixpkgs with niv in order to build everything. It won't work everywhere, however the hermetic flag in the deploy_nixos Terraform module will make this work.

Now let's deploy all this and see if it works!

$ terraform init

$ terraform plan

$ terraform apply

Printerfacts Install

Now we can add the printerfacts service to the VM. First, add the printerfacts repo to niv:

$ niv add git -n printerfacts --repo https://tulpa.dev/cadey/printerfacts
Done: Adding package printerfacts

Then create a service definition for it in your common folder. First create the folder common/services:

$ cd ../..
$ cd common
$ mkdir services
$ cd services

Then create a default.nix file with the following contents:

# common/services/default.nix

{ ... }:

{
  imports = [ ./printerfacts.nix ];
}

And create ./printerfacts.nix with this service boilerplate:

# common/services/printerfacts.nix

{ config, pkgs, lib, ... }:

with lib;

let
  sources = import ../../nix/sources.nix;
  pkg = pkgs.callPackage sources.printerfacts { };
  cfg = config.xeserv.services.printerfacts;
in
{
  options.xeserv.services.printerfacts = {
    enable = mkEnableOption "enable Printerfacts";
    useACME = mkEnableOption "enable ACME certs";

    domain = mkOption {
      type = types.str;
      default = "printerfacts.akua";
      example = "printerfacts.cetacean.club";
      description =
        "The domain name that nginx should check against for HTTP hostnames";
    };

    port = mkOption {
      type = types.int;
      default = 28318;
      example = 9001;
      description =
        "The port number printerfacts should listen on for HTTP traffic";
    };
  };

  config = mkIf cfg.enable {
    systemd.services.printerfacts = {
      wantedBy = [ "multi-user.target" ];

      script = ''
        export PORT=${toString cfg.port}
        export DOMAIN=${toString cfg.domain}
        export RUST_LOG=info
        exec ${pkg}/bin/printerfacts
      '';

      serviceConfig = {
        Restart = "always";
        RestartSec = "30s";
        WorkingDirectory = "${pkg}";
        RuntimeDirectory = "printerfacts";
        RuntimeDirectoryMode = "0755";
        StateDirectory = "tailscale";
        StateDirectoryMode = "0750";
        CacheDirectory = "tailscale";
        CacheDirectoryMode = "0750";
        DynamicUser = "yes";
      };
    };

    services.nginx.virtualHosts."${cfg.domain}" = {
      locations."/" = {
        proxyPass = "http://127.0.0.1:${toString cfg.port}";
        proxyWebsockets = true;
      };
      enableACME = cfg.useACME;
    };
  };
}

Then wire up common/default.nix with this:

# common/default.nix

imports = [ ./paranoid.nix ./users.nix ./services ];

Then you can add this to your machine config in the terraform directory:

# terraform/printerfacts/printerfacts.nix

configuration = { config, lib, pkgs, ... }: {
  # ...

  xeserv.services.printerfacts = {
    enable = true;
    domain = "3.237.88.228"; # replace this with the IP of your AWS instance
  };
};

Then terraform plan and terraform apply:

$ terraform plan

$ terraform apply

And finally get yourself a hard-earned printer fact:

$ curl http://3.237.88.228/fact
In 1987 printers overtook scanners as the number one pet in America.

We have gone from nothing to a fully production-ready NixOS deployment including a custom AMI pinned to an exact version of NixOS and an additional service added from its git repo. This will allow you to create a NixOS deployment that can be used by multiple people but will also stay pinned to an exact version of NixOS. Terraform will do all of the NixOS building and ensure that things are kept up to date, meaning that your infrastructure is all configured using the same workflow.

This post outlines boilerplate and templates. I'm sure that you could easily adapt these templates for other things as well. If you need to store persistent data, make sure its being put in /var/lib so that it isn't wiped on reboot. This took at least a week of research, banging my head against the wall and so many failures to implement this. Many thanks to Graham Christensen for unblocking me on this and pulling me back from the chasm a few times.

Hope this helps your prod NixOS adventures!

Facts and circumstances may have changed since publication. Please contact me before jumping to conclusions if something seems wrong or unclear.

Tags: paranix, aws, r13y