modules/private/monitoring/objects_common.nix

   1 { hostFQDN
   2 , hostName
   3 , interface ? "eth0"
   4 , processWarn ? "250"
   5 , processAlert ? "400"
   6 , loadWarn ? "0.9"
   7 , load5Warn ? loadWarn
   8 , load15Warn ? load5Warn
   9 , loadAlert ? "1.0"
  10 , load5Alert ? loadAlert
  11 , load15Alert ? load5Alert
  12 , mdadm
  13 , master
  14 , lib
  15 , mypluginsConfig
  16 , ...
  17 }:
  18 let
  19   defaultPassiveInfo = {
  20     filter = lib.attrsets.filterAttrs
  21       (k: v: builtins.elem k ["service_description"] || builtins.substring 0 1 k == "_");
  22     use = "external-passive-service";
  23     freshness_threshold = "450";
  24     retry_interval = "1";
  25     servicegroups = "webstatus-resources";
  26     host_name = hostFQDN;
  27   };
  28 in
  29 {
  30   host = {
  31     "${hostFQDN}" = {
  32       alias = hostFQDN;
  33       address = hostFQDN;
  34       use = "linux-server";
  35       hostgroups = "webstatus-hosts";
  36       _webstatus_name = hostName;
  37       _webstatus_vhost = "status.immae.eu";
  38     };
  39   };
  40   service = [
  41     {
  42       passiveInfo = defaultPassiveInfo;
  43       service_description = "Size on root partition";
  44       use = "local-service";
  45       check_command = ["check_local_disk" "20%" "10%" "/"];
  46     }
  47     {
  48       passiveInfo = defaultPassiveInfo;
  49       service_description = "Total number of process";
  50       use = "local-service";
  51       check_command = [
  52         "check_local_procs"
  53         processWarn
  54         processAlert
  55         "RSZDT"
  56       ];
  57     }
  58     {
  59       passiveInfo = defaultPassiveInfo;
  60       service_description = "Network bandwidth";
  61       use = "local-service";
  62       check_interval = "2";
  63       max_check_attempts = "20";
  64       retry_interval = "2";
  65       check_command = [
  66         "check_local_bandwidth"
  67         interface
  68         "20480" # kb/s
  69         "51200" # kb/s
  70       ];
  71     }
  72     {
  73       passiveInfo = defaultPassiveInfo;
  74       service_description = "Average load";
  75       use = "local-service";
  76       check_command = [
  77         "check_local_load"
  78         "${loadWarn},${load5Warn},${load15Warn}"
  79         "${loadAlert},${load5Alert},${load15Alert}"
  80       ];
  81     }
  82     {
  83       passiveInfo = defaultPassiveInfo;
  84       service_description = "Swap usage";
  85       use = "local-service";
  86       check_command = ["check_local_swap" "20" "10"];
  87     }
  88     {
  89       passiveInfo = defaultPassiveInfo;
  90       service_description = "Memory usage";
  91       use = "local-service";
  92       check_command = ["check_memory" "80" "90"];
  93     }
  94     {
  95       passiveInfo = defaultPassiveInfo;
  96       service_description = "NTP is activated and working";
  97       use = "local-service";
  98       check_command = ["check_ntp"];
  99     }
 100     {
 101       passiveInfo = defaultPassiveInfo;
 102       service_description = "No mdadm array is degraded";
 103       use = "local-service";
 104       check_command = [
 105         "check_command_output"
 106         "${mdadm}/bin/mdadm --monitor --scan -1"
 107         "^$"
 108         "-s 0 -r root"
 109       ];
 110     }
 111   ];
 112   command = lib.foldr (v: o: o // (v.commands or {})) {} (builtins.attrValues mypluginsConfig) // {
 113     check_local_disk = "$USER1$/check_disk -w $ARG1$ -c $ARG2$ -p $ARG3$";
 114     check_local_procs = "$USER1$/check_procs -w $ARG1$ -c $ARG2$ -s $ARG3$";
 115     check_local_load = "$USER1$/check_load -r -w $ARG1$ -c $ARG2$";
 116     check_local_swap = "$USER1$/check_swap -n ok -w $ARG1$ -c $ARG2$";
 117     check_ntp = "$USER1$/check_ntp_time -t 30 -q -H 0.arch.pool.ntp.org";
 118     check_mailq = "$USER1$/check_mailq -s -w 1 -c 2";
 119     check_smtp = "$USER1$/check_smtp -H $HOSTADDRESS$ -p 25 -S -D 21,15";
 120
 121     check_host_alive = "$USER1$/check_ping -H $HOSTADDRESS$ -w 3000.0,80% -c 5000.0,100% -p 5";
 122     check_ok = "$USER1$/check_dummy 0 \"Dummy OK\"";
 123     check_critical = "$USER1$/check_dummy 2 \"Dummy CRITICAL\"";
 124   };
 125   timeperiod = {
 126     "24x7" = {
 127       alias = "24 Hours A Day, 7 Days A Week";
 128       monday = "00:00-24:00";
 129       tuesday = "00:00-24:00";
 130       wednesday = "00:00-24:00";
 131       thursday = "00:00-24:00";
 132       friday = "00:00-24:00";
 133       saturday = "00:00-24:00";
 134       sunday = "00:00-24:00";
 135     };
 136   };
 137   servicegroup = {
 138     webstatus-webapps = { alias = "Web applications"; };
 139     webstatus-websites = { alias = "Personal websites"; };
 140     webstatus-ssl = { alias = "SSL certificates"; };
 141     webstatus-dns = { alias = "DNS resolution"; };
 142     webstatus-remote-services = { alias = "Other remote services"; };
 143     webstatus-local-services = { alias = "Other local services"; };
 144     webstatus-email = { alias = "E-mail services"; };
 145     webstatus-resources = { alias = "Local resources"; };
 146     webstatus-databases = { alias = "Databases resources"; };
 147     webstatus-backup = { alias = "Backup resources"; };
 148   };
 149   hostgroup = {
 150     webstatus-hosts   = { alias = "Hosts"; };
 151   };
 152   contactgroup = {
 153     admins = { alias = "Naemon Administrators"; };
 154   };
 155   templates = {
 156     service = {
 157       generic-service = {
 158         active_checks_enabled = "1";
 159         check_freshness = "0";
 160         check_interval = "10";
 161         check_period = "24x7";
 162         contact_groups = "admins";
 163         event_handler_enabled = "1";
 164         flap_detection_enabled = "1";
 165         is_volatile = "0";
 166         max_check_attempts = "3";
 167         notification_interval = "60";
 168         notification_options = "w,u,c,r,f,s";
 169         notification_period = "24x7";
 170         notifications_enabled = if master then "1" else "0";
 171         obsess_over_service = "1";
 172         passive_checks_enabled = "1";
 173         process_perf_data = "1";
 174         retain_nonstatus_information = "1";
 175         retain_status_information = "1";
 176         retry_interval = "2";
 177         _webstatus_namespace = "immae";
 178       };
 179       local-service = {
 180         use = "generic-service";
 181         host_name = hostFQDN;
 182         check_interval = "5";
 183         max_check_attempts = "4";
 184         retry_interval = "1";
 185         servicegroups = "webstatus-resources";
 186       };
 187       external-service = {
 188         use = "generic-service";
 189         check_interval = "5";
 190         max_check_attempts = "4";
 191         retry_interval = "1";
 192       };
 193       web-service = {
 194         use = "generic-service";
 195         check_interval = "20";
 196         max_check_attempts = "2";
 197         retry_interval = "1";
 198       };
 199       external-web-service = {
 200         use = "generic-service";
 201         check_interval = "20";
 202         max_check_attempts = "2";
 203         retry_interval = "1";
 204       };
 205       mail-service = {
 206         use = "generic-service";
 207         check_interval = "15";
 208         max_check_attempts = "1";
 209         retry_interval = "1";
 210       };
 211       dns-service = {
 212         use = "generic-service";
 213         check_interval = "120";
 214         notification_interval = "120";
 215         max_check_attempts = "5";
 216         retry_interval = "5";
 217       };
 218     };
 219     # No contact, we go through master
 220     contact = {
 221       generic-contact = {
 222         host_notification_commands = "notify-host-by-email";
 223         host_notification_options = "d,u,r,f,s";
 224         host_notification_period = "24x7";
 225         service_notification_commands = "notify-service-by-email";
 226         service_notification_options = "w,u,c,r,f,s";
 227         service_notification_period = "24x7";
 228       };
 229     };
 230     host = {
 231       generic-host = {
 232         event_handler_enabled = "1";
 233         flap_detection_enabled = "1";
 234         notification_period = "24x7";
 235         notifications_enabled = "1";
 236         process_perf_data = "1";
 237         retain_nonstatus_information = "1";
 238         retain_status_information = "1";
 239       };
 240       linux-server = {
 241         check_command = "check_host_alive";
 242         check_interval = "5";
 243         check_period = "24x7";
 244         contact_groups = "admins";
 245         max_check_attempts = "10";
 246         notification_interval = "120";
 247         notification_options = "d,u,r,f";
 248         retry_interval = "1";
 249         _webstatus_namespace = "immae";
 250       };
 251     };
 252   };
 253 }