nixos/tests/nagios.nix

   1 import ./make-test-python.nix (
   2   { pkgs, ... }: {
   3     name = "nagios";
   4     meta = with pkgs.lib.maintainers; {
   5       maintainers = [ symphorien ];
   6     };
   7
   8     nodes.machine = { lib, ... }: let
   9       writer = pkgs.writeShellScript "write" ''
  10         set -x
  11         echo "$@"  >> /tmp/notifications
  12       '';
  13     in
  14       {
  15         # tested service
  16         services.sshd.enable = true;
  17         # nagios
  18         services.nagios = {
  19           enable = true;
  20           # make state transitions faster
  21           extraConfig.interval_length = "5";
  22           objectDefs =
  23             (map (x: "${pkgs.nagios}/etc/objects/${x}.cfg") [ "templates" "timeperiods" "commands" ]) ++ [
  24               (
  25                 pkgs.writeText "objects.cfg" ''
  26                   # notifications are written to /tmp/notifications
  27                   define command {
  28                   command_name notify-host-by-file
  29                   command_line ${writer} "$HOSTNAME is $HOSTSTATE$"
  30                   }
  31                   define command {
  32                   command_name notify-service-by-file
  33                   command_line ${writer} "$SERVICEDESC$ is $SERVICESTATE$"
  34                   }
  35
  36                   # nagios boilerplate
  37                   define contact {
  38                   contact_name                    alice
  39                   alias                           alice
  40                   host_notifications_enabled      1
  41                   service_notifications_enabled   1
  42                   service_notification_period     24x7
  43                   host_notification_period        24x7
  44                   service_notification_options    w,u,c,r,f,s
  45                   host_notification_options       d,u,r,f,s
  46                   service_notification_commands   notify-service-by-file
  47                   host_notification_commands      notify-host-by-file
  48                   email                           foo@example.com
  49                   }
  50                   define contactgroup {
  51                   contactgroup_name   admins
  52                   alias               Admins
  53                   members alice
  54                   }
  55                   define hostgroup{
  56                   hostgroup_name  allhosts
  57                   alias  All hosts
  58                   }
  59
  60                   # monitored objects
  61                   define host {
  62                   use         generic-host
  63                   host_name   localhost
  64                   alias       localhost
  65                   address     localhost
  66                   hostgroups  allhosts
  67                   contact_groups admins
  68                   # make state transitions faster.
  69                   max_check_attempts 2
  70                   check_interval 1
  71                   retry_interval 1
  72                   }
  73                   define service {
  74                   use                 generic-service
  75                   host_name           localhost
  76                   service_description ssh
  77                   check_command       check_ssh
  78                   # make state transitions faster.
  79                   max_check_attempts 2
  80                   check_interval 1
  81                   retry_interval 1
  82                   }
  83                 ''
  84               )
  85             ];
  86         };
  87       };
  88
  89     testScript = { ... }: ''
  90       with subtest("ensure sshd starts"):
  91           machine.wait_for_unit("sshd.service")
  92
  93
  94       with subtest("ensure nagios starts"):
  95           machine.wait_for_file("/var/log/nagios/current")
  96
  97
  98       def assert_notify(text):
  99           machine.wait_for_file("/tmp/notifications")
 100           real = machine.succeed("cat /tmp/notifications").strip()
 101           print(f"got {real!r}, expected {text!r}")
 102           assert text == real
 103
 104
 105       with subtest("ensure we get a notification when sshd is down"):
 106           machine.succeed("systemctl stop sshd")
 107           assert_notify("ssh is CRITICAL")
 108
 109
 110       with subtest("ensure tests can succeed"):
 111           machine.succeed("systemctl start sshd")
 112           machine.succeed("rm /tmp/notifications")
 113           assert_notify("ssh is OK")
 114     '';
 115   }
 116 )