Put longer time for ntp checks

[perso/Immae/Projets/Puppet.git] / modules / profile / manifests / monitoring / services.pp

diff --git a/modules/profile/manifests/monitoring/services.pp b/modules/profile/manifests/monitoring/services.pp
index 39c2deff73194733f6169153b2d65b272593f7f6..5ff3c6dcb358215d3c367ed513f423535d212535 100644 (file)
--- a/modules/profile/manifests/monitoring/services.pp
+++ b/modules/profile/manifests/monitoring/services.pp
@@ -1,89 +1,57 @@
 class profile::monitoring::services {
-  $real_hostname = lookup("base_installation::real_hostname")
-
-  Nagios_service {
-    ensure => "present",
-    owner  => "naemon",
-    group  => "naemon",
-    target => "/etc/naemon/objects.cfg",
-    notify => Service["naemon"],
-    before => Service["naemon"],
-  }
-
-  $service_generic = {
-    active_checks_enabled        => "1",
-    check_freshness              => "0",
-    check_interval               => "10",
-    check_period                 => "24x7",
-    contact_groups               => "admins",
-    event_handler_enabled        => "1",
-    flap_detection_enabled       => "1",
-    is_volatile                  => "0",
-    max_check_attempts           => "3",
-    notification_interval        => "60",
-    notification_options         => "w,u,c,r",
-    notification_period          => "24x7",
-    notifications_enabled        => "1",
-    obsess_over_service          => "1",
-    passive_checks_enabled       => "1",
-    process_perf_data            => "1",
-    retain_nonstatus_information => "1",
-    retain_status_information    => "1",
-    retry_interval               => "2",
-  }
 
+  $warn_1_load  = 1.5*$facts["processorcount"]
+  $warn_5_load  = $facts["processorcount"]
+  $warn_15_load = $facts["processorcount"]
+  $warn_load = join([$warn_1_load, $warn_5_load, $warn_15_load], ",")
 
-  $service_local = merge($service_generic, {
-    host_name          => $real_hostname,
-    check_interval     => "5",
-    max_check_attempts => "4",
-    retry_interval     => "1",
-    })
+  $max_1_load  = 2*$facts["processorcount"]
+  $max_5_load  = $facts["processorcount"]
+  $max_15_load = $facts["processorcount"]
+  $max_load = join([$max_1_load, $max_5_load, $max_15_load], ",")
 
-  sudo::conf {
-    default:
-      sudo_file_name => "naemon";
-    'naemon-fail2ban':
-      content => "naemon  ALL=(root) NOPASSWD: /usr/bin/fail2ban-client ping";
-  }
-
-  nagios_service {
-    default: * => $service_local;
+  profile::monitoring::local_service {
     "Size on root partition":
-      service_description => "Size on root partition",
-      check_command       => "check_local_disk!20%!10%!/";
+      local => {
+        check_command => "check_local_disk!20%!10%!/",
+      };
     "Total number of process":
-      service_description => "Total number of process",
-      check_command       => "check_local_procs!250!400!RSZDT";
+      local => {
+        check_command => "check_local_procs!50!100!RSZDT",
+      };
     "Average load":
-      service_description => "Average load",
-      check_command       => "check_local_load!8.0,8.0,8.0!10.0,10.0,10.0";
+      local => {
+        check_command => "check_local_load!$warn_load!$max_load",
+      };
     "Swap usage":
-      service_description => "Swap usage",
-      check_command       => "check_local_swap!20!10";
+      local => {
+        check_command => "check_local_swap!20!10",
+      };
     "fail2ban is active":
-      service_description => "fail2ban is active",
-      check_command       => "check_command_output!fail2ban-client ping!pong!-r root",
-      require             => Sudo::Conf["naemon-fail2ban"];
+      sudos   => {
+        "naemon-fail2ban" => "naemon  ALL=(root) NOPASSWD: /usr/bin/fail2ban-client ping",
+      },
+      local   => {
+        check_command => "check_command_output!fail2ban-client ping!pong!-r root",
+      };
     "NTP is activated and working":
-      service_description => "NTP is activated and working",
-      check_command       => "check_ntp";
+      master => {
+        check_interval      => "60",
+        freshness_threshold => Integer(60*60*1.5),
+      },
+      local  => {
+        check_command  => "check_ntp",
+        check_interval => "60"
+      };
+    "No mdadm array is degraded":
+      common => {
+        ensure => (find_file("/proc/mdstat") == undef) ? { true => "absent", default =>"present" },
+      },
+      local => {
+        check_command => "check_md_raid",
+      };
   }
 
-  if empty(find_file("/proc/mdstat")) {
-    nagios_service {
-      default: * => $service_local;
-      "No mdadm array is degraded":
-        ensure              => "absent",
-        service_description => "No mdadm array is degraded",
-        check_command       => "check_md_raid";
-    }
-  } else {
-    nagios_service {
-      default: * => $service_local;
-      "No mdadm array is degraded":
-        service_description => "No mdadm array is degraded",
-        check_command       => "check_md_raid";
-    }
-  }
+  Profile::Monitoring::Local_service <| |>
+  Profile::Monitoring::External_service <| |>
 }