data model: add TF test (#506)

add a data model deployment test by VM using opentofu.
builds upon #505.

Reviewed-on: Fediversity/Fediversity#506

.forgejo/workflows/ci.yaml

@@ -64,6 +64,12 @@ jobs:
       - uses: actions/checkout@v4
       - run: nix build .#checks.x86_64-linux.deployment-panel -L
 
+  check-deployment-model:
+    runs-on: native
+    steps:
+      - uses: actions/checkout@v4
+      - run: nix build .#checks.x86_64-linux.deployment-model-ssh -L
+
   check-deployment-model-ssh:
     runs-on: native
     steps:
@@ -76,6 +82,12 @@ jobs:
       - uses: actions/checkout@v4
       - run: nix build .#checks.x86_64-linux.deployment-model-nixops4 -L
 
+  check-deployment-model-tf:
+    runs-on: native
+    steps:
+      - uses: actions/checkout@v4
+      - run: nix build .#checks.x86_64-linux.deployment-model-tf -L
+
   ## NOTE: NixOps4 does not provide a good “dry run” mode, so we instead check
   ## proxies for resources, namely whether their `.#vmOptions.<machine>` and
   ## `.#nixosConfigurations.<machine>` outputs evaluate and build correctly, and

deployment/check/common/data-model.nix

@@ -183,6 +183,29 @@ let
                     };
                 };
             };
+            single-nixos-vm-tf = environment: {
+              resources."operator-environment".login-shell.username = "operator";
+              implementation =
+                {
+                  required-resources,
+                  deployment-name,
+                }:
+                {
+                  tf-host = {
+                    nixos-configuration = mkNixosConfiguration environment required-resources;
+                    system = targetSystem;
+                    ssh = {
+                      username = "root";
+                      host = nodeName;
+                      key-file = null;
+                      inherit sshOpts;
+                    };
+                    module = self;
+                    inherit args deployment-name;
+                    root-path = pathToRoot;
+                  };
+                };
+            };
           };
       };
       options = {
@@ -215,6 +238,17 @@ let
               configuration = config."example-configuration";
             };
           };
+        "tf-deployment" =
+          let
+            env = config.environments."single-nixos-vm-tf";
+          in
+          mkOption {
+            type = env.resource-mapping.output-type;
+            default = env.deployment {
+              deployment-name = "tf-deployment";
+              configuration = config."example-configuration";
+            };
+          };
       };
     }
   );

deployment/check/data-model-ssh/nixosTest.nix

@@ -10,10 +10,7 @@ let
     inherit pathToRoot pathFromRoot;
     nodeName = "ssh";
     targetSystem = system;
-    sshOpts = [
-      "ConnectTimeout=1"
-      "ServerAliveInterval=1"
-    ];
+    sshOpts = [ ];
   };
   deploy =
     (import ../common/data-model.nix {

deployment/check/data-model-tf/constants.nix

@@ -0,0 +1,11 @@
+{
+  targetMachines = [
+    "target"
+  ];
+  pathToRoot = builtins.path {
+    path = ../../..;
+    name = "root";
+  };
+  pathFromRoot = "/deployment/check/data-model-tf";
+  enableAcme = true;
+}

deployment/check/data-model-tf/default.nix

@@ -0,0 +1,21 @@
+{
+  runNixOSTest,
+  inputs,
+  sources,
+}:
+
+runNixOSTest {
+  imports = [
+    ../../data-model.nix
+    ../../function.nix
+    ../common/nixosTest.nix
+    ./nixosTest.nix
+  ];
+  _module.args = { inherit inputs sources; };
+  inherit (import ./constants.nix)
+    targetMachines
+    pathToRoot
+    pathFromRoot
+    enableAcme
+    ;
+}

deployment/check/data-model-tf/nixosTest.nix

@@ -0,0 +1,63 @@
+{
+  lib,
+  pkgs,
+  ...
+}:
+let
+  inherit (import ./constants.nix) pathToRoot pathFromRoot;
+  inherit (pkgs) system;
+  deployment-config = {
+    inherit pathToRoot pathFromRoot;
+    nodeName = "target";
+    targetSystem = system;
+    sshOpts = [ ];
+  };
+  deploy =
+    (import ../common/data-model.nix {
+      inherit system;
+      config = deployment-config;
+      # opt not to pass `inputs`, as we could only pass serializable arguments through to its self-call
+    })."tf-deployment".tf-host.run;
+in
+{
+  _class = "nixosTest";
+  imports = [
+    ../common/data-model-options.nix
+  ];
+
+  name = "deployment-model";
+  sourceFileset = lib.fileset.unions [
+    ../../run/tf-single-host/run.sh
+  ];
+
+  nodes.deployer =
+    { pkgs, ... }:
+    {
+      environment.systemPackages = with pkgs; [
+        (pkgs.callPackage ../../run/tf-single-host/tf.nix { })
+        jq
+        deploy
+      ];
+
+      # needed only when building from deployer
+      system.extraDependenciesFromModule =
+        { pkgs, ... }:
+        {
+          environment.systemPackages = with pkgs; [
+            hello
+          ];
+        };
+    };
+
+  extraTestScript = ''
+    with subtest("ssh: Check the status before deployment"):
+      target.fail("hello 1>&2")
+
+    with subtest("ssh: Run the deployment"):
+      deployer.succeed("""
+        ${lib.getExe deploy}
+      """)
+      target.wait_for_unit("multi-user.target")
+      target.succeed("su - operator -c hello 1>&2")
+  '';
+}

deployment/data-model.nix

@@ -159,6 +159,76 @@ let
       description = "A NixOps4 NixOS deployment. For an example, see https://github.com/nixops4/nixops4-nixos/blob/main/example/deployment.nix.";
       type = nixops4Deployment;
     };
+    tf-host = mkOption {
+      description = "A Terraform deployment by SSH to update a single existing NixOS host.";
+      type = submodule (tf-host: {
+        options = {
+          system = mkOption {
+            description = "The architecture of the system to deploy to.";
+            type = types.str;
+          };
+          inherit nixos-configuration;
+          ssh = host-ssh;
+          module = mkOption {
+            description = "The module to call to obtain the NixOS configuration from.";
+            type = types.str;
+          };
+          args = mkOption {
+            description = "The arguments with which to call the module to obtain the NixOS configuration.";
+            type = types.attrs;
+          };
+          deployment-name = mkOption {
+            description = "The name of the deployment for which to obtain the NixOS configuration.";
+            type = types.str;
+          };
+          root-path = mkOption {
+            description = "The path to the root of the repository.";
+            type = types.path;
+          };
+          run = mkOption {
+            type = types.package;
+            # error: The option `tf-deployment.tf-host.run' is read-only, but it's set multiple times.
+            # readOnly = true;
+            default =
+              let
+                inherit (tf-host.config)
+                  system
+                  ssh
+                  module
+                  args
+                  deployment-name
+                  root-path
+                  ;
+                inherit (ssh)
+                  host
+                  username
+                  key-file
+                  sshOpts
+                  ;
+                environment = {
+                  key_file = key-file;
+                  deployment_name = deployment-name;
+                  root_path = root-path;
+                  ssh_opts = sshOpts;
+                  inherit
+                    system
+                    host
+                    username
+                    module
+                    args
+                    ;
+                  deployment_type = "tf-host";
+                };
+                tf-env = pkgs.callPackage ./run/tf-single-host/tf-env.nix { };
+              in
+              pkgs.writeShellScriptBin "deploy-ssh.sh" ''
+                env ${toString (lib.mapAttrsToList (k: v: "TF_VAR_${k}=\"${toBash v}\"") environment)} \
+                tf_env=${tf-env} bash ./deployment/run/tf-single-host/run.sh
+              '';
+          };
+        };
+      });
+    };
   };
 in
 {
@@ -190,6 +260,7 @@ in
                         type = types.optionType;
                       };
                       # TODO(@fricklerhandwerk): we may want to make the function type explicit here: `application-resources -> resource-type`
+                      # and then also rename this to be consistent with the application's resource mapping
                       options.apply = mkOption {
                         description = "Apply the policy to a request";
                         type = functionTo policy.config.resource-type;

deployment/flake-part.nix

@@ -36,6 +36,11 @@
           inherit (pkgs.testers) runNixOSTest;
           inherit inputs sources;
         };
+
+        deployment-model-tf = import ./check/data-model-tf {
+         inherit (pkgs.testers) runNixOSTest;
+          inherit inputs sources;
+        };
       };
     };
 }

deployment/run/ssh-single-host/run.sh

@@ -39,7 +39,7 @@ NIX_SSHOPTS="${sshOpts[*]}" nix-copy-closure --to "$destination" "$outPath" --gz
 # shellcheck disable=SC2029
 ssh "${sshOpts[@]}" "$destination" "nix-env --profile /nix/var/nix/profiles/system --set $outPath"
 # shellcheck disable=SC2029
-output=$(ssh "${sshOpts[@]}" "$destination" "nohup $outPath/bin/switch-to-configuration switch &" 2>&1) || echo "status code: $?"
+output=$(ssh -o "ConnectTimeout=1" -o "ServerAliveInterval=1" "${sshOpts[@]}" "$destination" "nohup $outPath/bin/switch-to-configuration switch &" 2>&1) || echo "status code: $?"
 echo "output: $output"
 if [[ $output != *"Timeout, server $host not responding"* ]]; then
   echo "non-timeout error: $output"

deployment/run/tf-single-host/main.tf

@@ -0,0 +1,52 @@
+# hash of our code directory, used to trigger re-deploy
+# FIXME calculate separately to reduce false positives
+data "external" "hash" {
+  program = ["sh", "-c", "echo \"{\\\"hash\\\":\\\"$(nix-hash ../../..)\\\"}\""]
+}
+
+# TF resource to build and deploy NixOS instances.
+resource "terraform_data" "nixos" {
+
+  # trigger rebuild/deploy if (FIXME?) any potentially used config/code changed,
+  # preventing these (20+s, build being bottleneck) when nothing changed.
+  # terraform-nixos separates these to only deploy if instantiate changed,
+  # yet building even then - which may be not as bad using deploy on remote.
+  # having build/deploy one resource reflects wanting to prevent no-op rebuilds
+  # over preventing (with less false positives) no-op deployments,
+  # as i could not find a way to do prevent no-op rebuilds without merging them:
+  # - generic resources cannot have outputs, while we want info from the instantiation (unless built on host?).
+  # - `data` always runs, which is slow for deploy and especially build.
+  triggers_replace = [
+    data.external.hash.result,
+    var.host,
+    var.module,
+    var.args,
+    var.root_path,
+    var.deployment_type,
+  ]
+
+  provisioner "local-exec" {
+    # directory to run the script from. we use the TF project root dir,
+    # here as a path relative from where TF is run from,
+    # matching calling modules' expectations on config_nix locations.
+    # note that absolute paths can cause false positives in triggers,
+    # so are generally discouraged in TF.
+    working_dir = path.root
+    environment = {
+      system = var.system
+      username = var.username
+      host = var.host
+      module = var.module
+      host = var.host
+      args = var.args
+      key_file = var.key_file
+      deployment_name = var.deployment_name
+      root_path = var.root_path
+      ssh_opts = var.ssh_opts
+      deployment_type = var.deployment_type
+    }
+    # TODO: refactor back to command="ignoreme" interpreter=concat([]) to protect sensitive data from error logs?
+    # TODO: build on target?
+    command = "sh ../ssh-single-host/run.sh"
+  }
+}

deployment/run/tf-single-host/run.sh

@@ -0,0 +1,9 @@
+#! /usr/bin/env bash
+set -xeuo pipefail
+declare tf_env
+
+export TF_LOG=info
+
+cd "${tf_env}/deployment/run/tf-single-host"
+# parallelism=1: limit OOM risk
+tofu apply --auto-approve -lock=false -parallelism=1

deployment/run/tf-single-host/setup.nix

@@ -0,0 +1,16 @@
+{
+  pkgs,
+  lib,
+  sources,
+}:
+pkgs.writeScriptBin "setup" ''
+  set -xe
+  # calculated pins
+  echo '${lib.strings.toJSON sources}' > ./.npins.json
+  # generate TF lock for nix's TF providers
+  rm -rf .terraform/
+  rm -f .terraform.lock.hcl
+  # suppress warning on architecture-specific generated lock file:
+  # `Warning: Incomplete lock file information for providers`.
+  tofu init -input=false 1>/dev/null
+''

deployment/run/tf-single-host/tf-env.nix

@@ -0,0 +1,31 @@
+{
+  lib,
+  pkgs,
+  sources ? import ../../../npins,
+}:
+pkgs.stdenv.mkDerivation {
+  name = "tf-repo";
+  src =
+    with lib.fileset;
+    toSource {
+      root = ../../../.;
+      # don't copy ignored files
+      fileset = intersection (gitTracked ../../../.) ../../../.;
+    };
+  buildInputs = [
+    (pkgs.callPackage ./tf.nix { })
+    (pkgs.callPackage ./setup.nix { inherit sources; })
+  ];
+  buildPhase = ''
+    runHook preBuild
+    pushd deployment/run/tf-single-host
+    source setup
+    popd
+    runHook postBuild
+  '';
+  installPhase = ''
+    runHook preInstall
+    cp -r . $out
+    runHook postInstall
+  '';
+}

deployment/run/tf-single-host/tf.nix

@@ -0,0 +1,11 @@
+# FIXME: use overlays so this gets imported just once?
+{
+  pkgs,
+  ...
+}:
+let
+  tf = pkgs.opentofu;
+in
+tf.withPlugins (p: [
+  p.external
+])

deployment/run/tf-single-host/variables.tf

@@ -0,0 +1,54 @@
+variable "system" {
+  description = "The architecture of the system to deploy to."
+  type = string
+  default = "x86_64-linux"
+}
+
+variable "username" {
+  description = "the SSH user to use"
+  type = string
+  default = "root"
+}
+
+variable "host" {
+  description = "the host to access by SSH"
+  type = string
+}
+
+variable "module" {
+  description = "The module to call to obtain the NixOS configuration from."
+  type = string
+}
+
+variable "args" {
+  description = "The arguments with which to call the module to obtain the NixOS configuration."
+  type = string
+  default = "{}"
+}
+
+variable "key_file" {
+  description = "path to the user's SSH private key"
+  type = string
+}
+
+variable "deployment_name" {
+  description = "The name of the deployment for which to obtain the NixOS configuration."
+  type = string
+}
+
+variable "root_path" {
+  description = "The path to the root of the repository."
+  type = string
+}
+
+variable "ssh_opts" {
+  description = "Extra SSH options (`-o`) to use."
+  type = string
+  default = "[]"
+}
+
+variable "deployment_type" {
+  description = "A `deployment-type` from the Fediversity data model, for grabbing the desired NixOS configuration."
+  type = string
+  default = "tf-host"
+}