From eb321df10d26c38460225c1877a55fa2961c8e0d Mon Sep 17 00:00:00 2001
From: Qubasa <consulting@qube.email>
Date: Wed, 17 Sep 2025 21:18:29 +0200
Subject: [PATCH] clanServices: add indepth test for telegraf metrics

---
 clanServices/monitoring/flake-module.nix      |  2 +
 clanServices/monitoring/tests/vm/default.nix  | 69 ++++++++++++++++---
 .../tests/vm/sops/machines/peer1/key.json     |  2 +-
 3 files changed, 63 insertions(+), 10 deletions(-)

diff --git a/clanServices/monitoring/flake-module.nix b/clanServices/monitoring/flake-module.nix
index cf159aaf2..8ef1f262d 100644
--- a/clanServices/monitoring/flake-module.nix
+++ b/clanServices/monitoring/flake-module.nix
@@ -1,4 +1,6 @@
 {
+  lib,
+  self,
   ...
 }:
 let
diff --git a/clanServices/monitoring/tests/vm/default.nix b/clanServices/monitoring/tests/vm/default.nix
index 12f988698..8c0a4226c 100644
--- a/clanServices/monitoring/tests/vm/default.nix
+++ b/clanServices/monitoring/tests/vm/default.nix
@@ -5,7 +5,9 @@
   clan = {
     directory = ./.;
     inventory = {
-      machines.peer1 = { };
+      machines.peer1 = {
+        deploy.targetHost = "[2001:db8:1::1]";
+      };
 
       instances."test" = {
         module.name = "monitoring";
@@ -19,6 +21,20 @@
     };
   };
 
+  nodes = {
+    peer1 =
+      { lib, ... }:
+      {
+        services.telegraf.extraConfig = {
+          agent.interval = lib.mkForce "1s";
+          outputs.prometheus_client = {
+            basic_password = lib.mkForce "";
+            basic_username = lib.mkForce "";
+          };
+        };
+      };
+  };
+
   extraPythonPackages = _p: [
     (pkgs.python3.pkgs.toPythonModule packages.${pkgs.system}.clan-cli)
   ];
@@ -27,21 +43,56 @@
     { ... }:
     ''
       import time
+      import os
+      import sys
+      import subprocess 
+      import json
+      import shlex
+      import urllib.request
+      from base64 import b64encode
       start_all()
 
       peer1.wait_for_unit("network-online.target")
       peer1.wait_for_unit("telegraf.service")
-      time.sleep(99999)
-      # peer1.wait_for_unit("telegraf-json.service")
+      peer1.wait_for_unit("telegraf-json.service")
+      peer1.succeed("curl http://localhost:9990")
+      peer1.succeed("curl http://localhost:9273/metrics")
 
-      from clan_lib.metrics.version import get_nixos_systems
-      from clan_lib.machines.machines import Machine as ClanMachine
-      from clan_lib.flake import Flake
+      # Fetch the basic auth password from the secret file
+      password = peer1.succeed("cat /var/run/secrets/vars/telegraf/password")
+      url = f"http://192.168.1.1:9990"
+      credentials = f"prometheus:{password}"
+      print("Using credentials:", credentials)
+      time.sleep(10)  # wait a bit for telegraf to collect some data
 
-      # mymachine = ClanMachine("peer1", Flake("."))
-      # data = get_nixos_systems(mymachine, )
+      # Fetch the json output from miniserve
+      encoded_credentials = b64encode(credentials.encode("utf-8")).decode("utf-8")
+      headers = {"Authorization": f"Basic {encoded_credentials}"}
+      req = urllib.request.Request(url, headers=headers)  # noqa: S310
+      response = urllib.request.urlopen(req)
+
+      # Look for the nixos_systems metric in the json output
+      found_system = False
+      for line in response:
+        line_str = line.decode("utf-8").strip()
+        line = json.loads(line_str)
+        if line["name"] == "nixos_systems":
+            found_system = True
+            print("Found nixos_systems metric in json output")
+            break
+        print(line)
+      assert found_system, "nixos_systems metric not found in json output"
+
+      # TODO: I would like to test the python code here but it's not working yet
+      # Missing: I need a way to get the encrypted var from the clan
+      #from clan_lib.metrics.version import get_nixos_systems
+      #from clan_lib.machines.machines import Machine as ClanMachine
+      #from clan_lib.flake import Flake
+      #from clan_lib.ssh.remote import Remote
+      #target_host = Remote("peer1", "192.168.1.1")
+      #machine = ClanMachine("peer1", flake=Flake("${./.}"))
+      # data = get_nixos_systems(mymachine, target_host)
       # assert data["current_system"] is not None
 
-      time.sleep(99999)
     '';
 }
diff --git a/clanServices/monitoring/tests/vm/sops/machines/peer1/key.json b/clanServices/monitoring/tests/vm/sops/machines/peer1/key.json
index 09b67a652..bc3303787 100755
--- a/clanServices/monitoring/tests/vm/sops/machines/peer1/key.json
+++ b/clanServices/monitoring/tests/vm/sops/machines/peer1/key.json
@@ -3,4 +3,4 @@
     "publickey": "age18vspwr3de7jp0awyu66kk9psd5x4sy9suv0zt7ux2kqw0s6h2ueqwkgjxm",
     "type": "age"
   }
-]
\ No newline at end of file
+]