Merge pull request #76551 from symphorien/nagios2

nagios: various improvements

nixos/modules/services/monitoring/nagios.nix

@@ -17,32 +17,39 @@ let
       preferLocalBuild = true;
     } "mkdir -p $out; ln -s $nagiosObjectDefs $out/";
 
-  nagiosCfgFile = pkgs.writeText "nagios.cfg"
+  nagiosCfgFile = let
-    ''
+    default = {
-      # Paths for state and logs.
+      log_file="${nagiosLogDir}/current";
-      log_file=${nagiosLogDir}/current
+      log_archive_path="${nagiosLogDir}/archive";
-      log_archive_path=${nagiosLogDir}/archive
+      status_file="${nagiosState}/status.dat";
-      status_file=${nagiosState}/status.dat
+      object_cache_file="${nagiosState}/objects.cache";
-      object_cache_file=${nagiosState}/objects.cache
+      temp_file="${nagiosState}/nagios.tmp";
-      temp_file=${nagiosState}/nagios.tmp
+      lock_file="/run/nagios.lock";
-      lock_file=/run/nagios.lock # Not used I think.
+      state_retention_file="${nagiosState}/retention.dat";
-      state_retention_file=${nagiosState}/retention.dat
+      query_socket="${nagiosState}/nagios.qh";
-      query_socket=${nagiosState}/nagios.qh
+      check_result_path="${nagiosState}";
-      check_result_path=${nagiosState}
+      command_file="${nagiosState}/nagios.cmd";
-      command_file=${nagiosState}/nagios.cmd
+      cfg_dir="${nagiosObjectDefsDir}";
-
+      nagios_user="nagios";
-      # Configuration files.
+      nagios_group="nagios";
-      #resource_file=resource.cfg
+      illegal_macro_output_chars="`~$&|'\"<>";
-      cfg_dir=${nagiosObjectDefsDir}
+      retain_state_information="1";
-
+    };
-      # Uid/gid that the daemon runs under.
+    lines = mapAttrsToList (key: value: "${key}=${value}") (default // cfg.extraConfig);
-      nagios_user=nagios
+    content = concatStringsSep "\n" lines;
-      nagios_group=nagios
+    file = pkgs.writeText "nagios.cfg" content;
-
+    validated =  pkgs.runCommand "nagios-checked.cfg" {preferLocalBuild=true;} ''
-      # Misc. options.
+      cp ${file} nagios.cfg
-      illegal_macro_output_chars=`~$&|'"<>
+      # nagios checks the existence of /var/lib/nagios, but
-      retain_state_information=1
+      # it does not exists in the build sandbox, so we fake it
-    ''; # "
+      mkdir lib
+      lib=$(readlink -f lib)
+      sed -i s@=${nagiosState}@=$lib@ nagios.cfg
+      ${pkgs.nagios}/bin/nagios -v nagios.cfg && cp ${file} $out
+    '';
+    defaultCfgFile = if cfg.validateConfig then validated else file;
+  in
+  if cfg.mainConfigFile == null then defaultCfgFile else cfg.mainConfigFile;
 
   # Plain configuration for the Nagios web-interface with no
   # authentication.
@@ -77,16 +84,11 @@ in
     (mkRemovedOptionModule [ "services" "nagios" "urlPath" ] "The urlPath option has been removed as it is hard coded to /nagios in the nagios package.")
   ];
 
+  meta.maintainers = with lib.maintainers; [ symphorien ];
+
   options = {
     services.nagios = {
-      enable = mkOption {
+      enable = mkEnableOption "<link xlink:href='http://www.nagios.org/'>Nagios</link> to monitor your system or network.";
-        default = false;
-        description = "
-          Whether to use <link
-          xlink:href='http://www.nagios.org/'>Nagios</link> to monitor
-          your system or network.
-        ";
-      };
 
       objectDefs = mkOption {
         description = "
@@ -94,12 +96,14 @@ in
           the hosts, host groups, services and contacts for the
           network that you want Nagios to monitor.
         ";
+        type = types.listOf types.path;
+        example = literalExample "[ ./objects.cfg ]";
       };
 
       plugins = mkOption {
         type = types.listOf types.package;
-        default = [pkgs.nagiosPluginsOfficial pkgs.ssmtp];
+        default = with pkgs; [ nagiosPluginsOfficial ssmtp mailutils ];
-        defaultText = "[pkgs.nagiosPluginsOfficial pkgs.ssmtp]";
+        defaultText = "[pkgs.nagiosPluginsOfficial pkgs.ssmtp pkgs.mailutils]";
         description = "
           Packages to be added to the Nagios <envar>PATH</envar>.
           Typically used to add plugins, but can be anything.
@@ -107,14 +111,29 @@ in
       };
 
       mainConfigFile = mkOption {
-        type = types.package;
+        type = types.nullOr types.package;
-        default = nagiosCfgFile;
+        default = null;
-        defaultText = "nagiosCfgFile";
         description = "
-          Derivation for the main configuration file of Nagios.
+          If non-null, overrides the main configuration file of Nagios.
         ";
       };
 
+      extraConfig = mkOption {
+        type = types.attrsOf types.str;
+        example = {
+          debug_level = "-1";
+          debug_file = "/var/log/nagios/debug.log";
+        };
+        default = {};
+        description = "Configuration to add to /etc/nagios.cfg";
+      };
+
+      validateConfig = mkOption {
+        type = types.bool;
+        default = pkgs.stdenv.hostPlatform == pkgs.stdenv.buildPlatform;
+        description = "if true, the syntax of the nagios configuration file is checked at build time";
+      };
+
       cgiConfigFile = mkOption {
         type = types.package;
         default = nagiosCGICfgFile;
@@ -126,6 +145,7 @@ in
       };
 
       enableWebInterface = mkOption {
+        type = types.bool;
         default = false;
         description = "
           Whether to enable the Nagios web interface.  You should also
@@ -165,7 +185,7 @@ in
     # This isn't needed, it's just so that the user can type "nagiostats
     # -c /etc/nagios.cfg".
     environment.etc = [
-      { source = cfg.mainConfigFile;
+      { source = nagiosCfgFile;
         target = "nagios.cfg";
       }
     ];
@@ -173,7 +193,7 @@ in
     environment.systemPackages = [ pkgs.nagios ];
     systemd.services.nagios = {
       description = "Nagios monitoring daemon";
-      path     = [ pkgs.nagios ];
+      path     = [ pkgs.nagios ] ++ cfg.plugins;
       wantedBy = [ "multi-user.target" ];
       after    = [ "network.target" ];
 
@@ -184,14 +204,9 @@ in
         RestartSec = 2;
         LogsDirectory = "nagios";
         StateDirectory = "nagios";
+        ExecStart = "${pkgs.nagios}/bin/nagios /etc/nagios.cfg";
+        X-ReloadIfChanged = nagiosCfgFile;
       };
-
-      script = ''
-        for i in ${toString cfg.plugins}; do
-          export PATH=$i/bin:$i/sbin:$i/libexec:$PATH
-        done
-        exec ${pkgs.nagios}/bin/nagios ${cfg.mainConfigFile}
-      '';
     };
 
     services.httpd.virtualHosts = optionalAttrs cfg.enableWebInterface {

nixos/tests/all-tests.nix

@@ -179,6 +179,7 @@ in
   mysql = handleTest ./mysql.nix {};
   mysqlBackup = handleTest ./mysql-backup.nix {};
   mysqlReplication = handleTest ./mysql-replication.nix {};
+  nagios = handleTest ./nagios.nix {};
   nat.firewall = handleTest ./nat.nix { withFirewall = true; };
   nat.firewall-conntrack = handleTest ./nat.nix { withFirewall = true; withConntrackHelpers = true; };
   nat.standalone = handleTest ./nat.nix { withFirewall = false; };

nixos/tests/nagios.nix

@@ -0,0 +1,116 @@
+import ./make-test-python.nix (
+  { pkgs, ... }: {
+    name = "nagios";
+    meta = with pkgs.stdenv.lib.maintainers; {
+      maintainers = [ symphorien ];
+    };
+
+    machine = { lib, ... }: let
+      writer = pkgs.writeShellScript "write" ''
+        set -x
+        echo "$@"  >> /tmp/notifications
+      '';
+    in
+      {
+        # tested service
+        services.sshd.enable = true;
+        # nagios
+        services.nagios = {
+          enable = true;
+          # make state transitions faster
+          extraConfig.interval_length = "5";
+          objectDefs =
+            (map (x: "${pkgs.nagios}/etc/objects/${x}.cfg") [ "templates" "timeperiods" "commands" ]) ++ [
+              (
+                pkgs.writeText "objects.cfg" ''
+                  # notifications are written to /tmp/notifications
+                  define command {
+                  command_name notify-host-by-file
+                  command_line ${writer} "$HOSTNAME is $HOSTSTATE$"
+                  }
+                  define command {
+                  command_name notify-service-by-file
+                  command_line ${writer} "$SERVICEDESC$ is $SERVICESTATE$"
+                  }
+
+                  # nagios boilerplate
+                  define contact {
+                  contact_name                    alice
+                  alias                           alice
+                  host_notifications_enabled      1
+                  service_notifications_enabled   1
+                  service_notification_period     24x7
+                  host_notification_period        24x7
+                  service_notification_options    w,u,c,r,f,s
+                  host_notification_options       d,u,r,f,s
+                  service_notification_commands   notify-service-by-file
+                  host_notification_commands      notify-host-by-file
+                  email                           foo@example.com
+                  }
+                  define contactgroup {
+                  contactgroup_name   admins
+                  alias               Admins
+                  members alice
+                  }
+                  define hostgroup{
+                  hostgroup_name  allhosts
+                  alias  All hosts
+                  }
+
+                  # monitored objects
+                  define host {
+                  use         generic-host
+                  host_name   localhost
+                  alias       localhost
+                  address     localhost
+                  hostgroups  allhosts
+                  contact_groups admins
+                  # make state transitions faster.
+                  max_check_attempts 2
+                  check_interval 1
+                  retry_interval 1
+                  }
+                  define service {
+                  use                 generic-service
+                  host_name           localhost
+                  service_description ssh
+                  check_command       check_ssh
+                  # make state transitions faster.
+                  max_check_attempts 2
+                  check_interval 1
+                  retry_interval 1
+                  }
+                ''
+              )
+            ];
+        };
+      };
+
+    testScript = { ... }: ''
+      with subtest("ensure sshd starts"):
+          machine.wait_for_unit("sshd.service")
+
+
+      with subtest("ensure nagios starts"):
+          machine.wait_for_file("/var/log/nagios/current")
+
+
+      def assert_notify(text):
+          machine.wait_for_file("/tmp/notifications")
+          real = machine.succeed("cat /tmp/notifications").strip()
+          print(f"got {real!r}, expected {text!r}")
+          assert text == real
+
+
+      with subtest("ensure we get a notification when sshd is down"):
+          machine.succeed("systemctl stop sshd")
+          assert_notify("ssh is CRITICAL")
+
+
+      with subtest("ensure tests can succeed"):
+          machine.succeed("systemctl start sshd")
+          machine.succeed("rm /tmp/notifications")
+          assert_notify("ssh is OK")
+    '';
+  }
+)

pkgs/servers/monitoring/nagios/default.nix

@@ -1,4 +1,4 @@
-{ stdenv, fetchurl, perl, php, gd, libpng, zlib, unzip }:
+{ stdenv, fetchurl, perl, php, gd, libpng, zlib, unzip, nixosTests }:
 
 stdenv.mkDerivation rec {
   pname = "nagios";
@@ -20,6 +20,16 @@ stdenv.mkDerivation rec {
     substituteInPlace Makefile --replace '$(MAKE) install-basic' ""
   '';
   installTargets = "install install-config";
+  postInstall = ''
+    # don't make default files use hardcoded paths to commands
+    sed -i 's@command_line *[^ ]*/\([^/]*\) @command_line \1 @'  $out/etc/objects/commands.cfg
+    sed -i 's@/usr/bin/@@g' $out/etc/objects/commands.cfg
+    sed -i 's@/bin/@@g' $out/etc/objects/commands.cfg
+  '';
+
+  passthru.tests = {
+    inherit (nixosTests) nagios;
+  };
 
   meta = {
     description = "A host, service and network monitoring program";