modules/private/monitoring/objects_common.nix

   1 { hostFQDN
   2 , hostName
   3 , interface ? "eth0"
   4 , processWarn ? "250"
   5 , processAlert ? "400"
   6 , loadWarn ? "8.0"
   7 , load5Warn ? loadWarn
   8 , load15Warn ? load5Warn
   9 , loadAlert ? "10.0"
  10 , load5Alert ? loadAlert
  11 , load15Alert ? load5Alert
  12 , mdadm
  13 , sudo
  14 , master
  15 , lib
  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         "^$"
 108         "-s 0 -r root"
 109       ];
 110     }
 111   ];
 112   command = {
 113     check_dns = "$USER1$/check_dns -H $ARG1$ -s $HOSTADDRESS$ $ARG2$";
 114     check_emails = "$USER2$/check_emails -H $HOSTADDRESS$ -i $USER203$ -l $ARG1$ -p $ARG2$ -s $ARG3$ -f $ARG4$";
 115     check_emails_local = "$USER2$/check_emails -H $HOSTADDRESS$ -n $ARG1$ -r $ADMINEMAIL$ -s $ARG2$ -f $ARG3$";
 116     check_backup_eriomem = "$USER2$/check_eriomem $USER208$";
 117     check_backup_eriomem_age = "$USER2$/check_backup_eriomem_age $ARG1$";
 118     check_backup_ovh_age = "$USER2$/check_backup_ovh_age $ARG1$";
 119     check_external_dns = "$USER1$/check_dns -H $ARG2$ -s $ARG1$ $ARG3$";
 120     check_ftp_database = "$USER2$/check_ftp_database";
 121     check_git = "$USER2$/check_git $USER203$";
 122     check_http = "$USER1$/check_http --sni -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -r \"$ARG3$\"";
 123     check_https = "$USER1$/check_http --sni --ssl -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -r \"$ARG3$\"";
 124     check_https_auth = "$USER1$/check_http --sni --ssl -a \"$USER202$\" -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -r \"$ARG3$\"";
 125     check_https_certificate = "$USER1$/check_http --sni --ssl -H \"$ARG1$\" -C 21,15";
 126     check_https_code = "$USER1$/check_http --sni --ssl -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -e \"$ARG3$\" -r \"$ARG4$\"";
 127     check_imap_connection = "$USER2$/check_imap_connection -u \"$USER204$\" -p \"$USER205$\" -H \"imap.immae.eu:143\"";
 128     check_local_disk = "$USER1$/check_disk -w $ARG1$ -c $ARG2$ -p $ARG3$";
 129     check_local_procs = "$USER1$/check_procs -w $ARG1$ -c $ARG2$ -s $ARG3$";
 130     check_local_load = "$USER1$/check_load -w $ARG1$ -c $ARG2$";
 131     check_local_swap = "$USER1$/check_swap -n ok -w $ARG1$ -c $ARG2$";
 132     check_local_bandwidth = "$USER2$/check_bandwidth -i=$ARG1$ -w $ARG2$ -c $ARG3$";
 133     check_memory = "$USER2$/check_mem.sh -w $ARG1$ -c $ARG2$";
 134     check_command_match = "$USER2$/check_command -c \"$ARG1$\" -C \"$ARG2$\" $ARG3$";
 135     check_command_output = "$USER2$/check_command -c \"$ARG1$\" -s 0 -o \"$ARG2$\" $ARG3$";
 136     check_command_status = "$USER2$/check_command -c \"$ARG1$\" -s \"$ARG2$\" $ARG3$";
 137     check_ntp = "$USER1$/check_ntp_time -t 30 -q -H 0.arch.pool.ntp.org";
 138     check_mailq = "$USER1$/check_mailq -s -w 1 -c 2";
 139     check_megaraid = "$USER2$/check_megaraid_sas --sudo";
 140     check_mysql_replication = "${sudo} -u mysql $USER2$/check_mysql_replication \"$ARG1$\" \"$ARG2$\"";
 141     check_postgresql_replication = "${sudo} -u postgres $USER2$/check_postgres_replication \"$ARG1$\" \"$ARG2$\" \"$ARG3$\"";
 142     check_openldap_replication = "${sudo} -u openldap $USER2$/check_openldap_replication \"$ARG1$\" \"$ARG2$\" \"$ARG3$\" \"$ARG4$\" \"$ARG5$\"";
 143     check_ovh_sms = "$USER2$/check_ovh_sms \"$USER209$\"";
 144     check_redis_replication = "${sudo} -u redis $USER2$/check_redis_replication \"$ARG1$\"";
 145     check_smtp = "$USER1$/check_smtp -H $HOSTADDRESS$ -p 25 -S -D 21,15";
 146     check_tcp = "$USER1$/check_tcp -H $HOSTADDRESS$ -p $ARG1$ -e \"$ARG2$\" -Mcrit";
 147     check_tcp_ssl = "$USER1$/check_tcp -H $HOSTADDRESS$ -p $ARG1$ -S -D 21,15";
 148     check_zfs = "$USER2$/check_zpool.sh -p ALL";
 149
 150     check_host_alive = "$USER1$/check_ping -H $HOSTADDRESS$ -w 3000.0,80% -c 5000.0,100% -p 5";
 151     check_last_file_date = "${sudo} -u \"$ARG3$\" $USER2$/check_last_file_date \"$ARG1$\" \"$ARG2$\"";
 152     check_ok = "$USER1$/check_dummy 0 \"Dummy OK\"";
 153     check_critical = "$USER1$/check_dummy 2 \"Dummy CRITICAL\"";
 154
 155     # $OVE is to force naemon to run via shell instead of execve which fails here
 156     notify-host-by-email = "ADMINEMAIL=\"$ADMINEMAIL$\" SERVICENOTIFICATIONID=\"$SERVICENOTIFICATIONID$\" HOSTSTATE=\"$HOSTSTATE$\" HOSTOUTPUT=\"$HOSTOUTPUT$\" $USER2$/notify_by_email host \"$NOTIFICATIONTYPE$\" \"$HOSTALIAS$\" \"$LONGDATETIME$\" \"$CONTACTEMAIL$\" $OVE";
 157     # $OVE is to force naemon to run via shell instead of execve which fails here
 158     notify-service-by-email = "ADMINEMAIL=\"$ADMINEMAIL$\" SERVICENOTIFICATIONID=\"$SERVICENOTIFICATIONID$\" SERVICEDESC=\"$SERVICEDESC$\" SERVICESTATE=\"$SERVICESTATE$\" SERVICEOUTPUT=\"$SERVICEOUTPUT$\" $USER2$/notify_by_email service \"$NOTIFICATIONTYPE$\" \"$HOSTALIAS$\" \"$LONGDATETIME$\" \"$CONTACTEMAIL$\" $OVE";
 159     notify-by-slack = "HOST=\"$HOSTALIAS$\" SERVICESTATE=\"$SERVICESTATE$\" SERVICEDESC=\"$SERVICEDESC$\" SERVICEOUTPUT=\"$SERVICEOUTPUT$\" $USER2$/notify_by_slack \"$ARG1$\" \"$ARG2$\"";
 160     notify-host-eban-url = "STATUS_NAME=\"Server\" PASSWORD=\"$USER210$\" HOSTSTATE=\"$HOSTSTATE$\" $USER2$/notify_eban_url";
 161     notify-service-eban-url = "STATUS_NAME=\"$_SERVICEWEBSTATUS_NAME$\" PASSWORD=\"$USER210$\" SERVICESTATE=\"$SERVICESTATE$\" $USER2$/notify_eban_url";
 162
 163     notify-master = "$USER2$/send_nrdp.sh -u \"$USER200$\" -t \"$USER201$\" -H \"$HOSTADDRESS$\" -s \"$SERVICEDESC$\" -S \"$SERVICESTATEID$\" -o \"$SERVICEOUTPUT$ | $SERVICEPERFDATA$\"";
 164   };
 165   timeperiod = {
 166     "24x7" = {
 167       alias = "24 Hours A Day, 7 Days A Week";
 168       monday = "00:00-24:00";
 169       tuesday = "00:00-24:00";
 170       wednesday = "00:00-24:00";
 171       thursday = "00:00-24:00";
 172       friday = "00:00-24:00";
 173       saturday = "00:00-24:00";
 174       sunday = "00:00-24:00";
 175     };
 176   };
 177   servicegroup = {
 178     webstatus-webapps = { alias = "Web applications"; };
 179     webstatus-websites = { alias = "Personal websites"; };
 180     webstatus-ssl = { alias = "SSL certificates"; };
 181     webstatus-dns = { alias = "DNS resolution"; };
 182     webstatus-remote-services = { alias = "Other remote services"; };
 183     webstatus-local-services = { alias = "Other local services"; };
 184     webstatus-email = { alias = "E-mail services"; };
 185     webstatus-resources = { alias = "Local resources"; };
 186     webstatus-databases = { alias = "Databases resources"; };
 187     webstatus-backup = { alias = "Backup resources"; };
 188   };
 189   hostgroup = {
 190     webstatus-hosts = { alias = "Hosts"; };
 191   };
 192   contactgroup = {
 193     admins = { alias = "Naemon Administrators"; };
 194   };
 195   templates = {
 196     service = {
 197       generic-service = {
 198         active_checks_enabled = "1";
 199         check_freshness = "0";
 200         check_interval = "10";
 201         check_period = "24x7";
 202         contact_groups = "admins";
 203         event_handler_enabled = "1";
 204         flap_detection_enabled = "1";
 205         is_volatile = "0";
 206         max_check_attempts = "3";
 207         notification_interval = "60";
 208         notification_options = "w,u,c,r,f,s";
 209         notification_period = "24x7";
 210         notifications_enabled = if master then "1" else "0";
 211         obsess_over_service = "1";
 212         passive_checks_enabled = "1";
 213         process_perf_data = "1";
 214         retain_nonstatus_information = "1";
 215         retain_status_information = "1";
 216         retry_interval = "2";
 217       };
 218       local-service = {
 219         use = "generic-service";
 220         host_name = hostFQDN;
 221         check_interval = "5";
 222         max_check_attempts = "4";
 223         retry_interval = "1";
 224         servicegroups = "webstatus-resources";
 225       };
 226       external-service = {
 227         use = "generic-service";
 228         check_interval = "5";
 229         max_check_attempts = "4";
 230         retry_interval = "1";
 231       };
 232       web-service = {
 233         use = "generic-service";
 234         check_interval = "20";
 235         max_check_attempts = "2";
 236         retry_interval = "1";
 237       };
 238       external-web-service = {
 239         use = "generic-service";
 240         check_interval = "20";
 241         max_check_attempts = "2";
 242         retry_interval = "1";
 243       };
 244       mail-service = {
 245         use = "generic-service";
 246         check_interval = "15";
 247         max_check_attempts = "1";
 248         retry_interval = "1";
 249       };
 250       dns-service = {
 251         use = "generic-service";
 252         check_interval = "120";
 253         notification_interval = "120";
 254         max_check_attempts = "5";
 255         retry_interval = "5";
 256       };
 257     };
 258     # No contact, we go through master
 259     contact = {
 260       generic-contact = {
 261         host_notification_commands = "notify-host-by-email";
 262         host_notification_options = "d,u,r,f,s";
 263         host_notification_period = "24x7";
 264         service_notification_commands = "notify-service-by-email";
 265         service_notification_options = "w,u,c,r,f,s";
 266         service_notification_period = "24x7";
 267       };
 268     };
 269     host = {
 270       generic-host = {
 271         event_handler_enabled = "1";
 272         flap_detection_enabled = "1";
 273         notification_period = "24x7";
 274         notifications_enabled = "1";
 275         process_perf_data = "1";
 276         retain_nonstatus_information = "1";
 277         retain_status_information = "1";
 278       };
 279       linux-server = {
 280         check_command = "check_host_alive";
 281         check_interval = "5";
 282         check_period = "24x7";
 283         contact_groups = "admins";
 284         max_check_attempts = "10";
 285         notification_interval = "120";
 286         notification_options = "d,u,r,f";
 287         retry_interval = "1";
 288       };
 289     };
 290   };
 291 }