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     "${hostFQDN}" = {
  32       alias = hostFQDN;
  33       address = hostFQDN;
  34       use = "linux-server";
  35       hostgroups = "webstatus-hosts";
  36       _webstatus_name = hostName;
  37     };
  38   };
  39   service = [
  40     {
  41       passiveInfo = defaultPassiveInfo;
  42       service_description = "Size on root partition";
  43       use = "local-service";
  44       check_command = ["check_local_disk" "20%" "10%" "/"];
  45     }
  46     {
  47       passiveInfo = defaultPassiveInfo;
  48       service_description = "Total number of process";
  49       use = "local-service";
  50       check_command = [
  51         "check_local_procs"
  52         processWarn
  53         processAlert
  54         "RSZDT"
  55       ];
  56     }
  57     {
  58       passiveInfo = defaultPassiveInfo;
  59       service_description = "Network bandwidth";
  60       use = "local-service";
  61       check_interval = "2";
  62       max_check_attempts = "20";
  63       retry_interval = "2";
  64       check_command = [
  65         "check_local_bandwidth"
  66         interface
  67         "20480" # kb/s
  68         "51200" # kb/s
  69       ];
  70     }
  71     {
  72       passiveInfo = defaultPassiveInfo;
  73       service_description = "Average load";
  74       use = "local-service";
  75       check_command = [
  76         "check_local_load"
  77         "${loadWarn},${load5Warn},${load15Warn}"
  78         "${loadAlert},${load5Alert},${load15Alert}"
  79       ];
  80     }
  81     {
  82       passiveInfo = defaultPassiveInfo;
  83       service_description = "Swap usage";
  84       use = "local-service";
  85       check_command = ["check_local_swap" "20" "10"];
  86     }
  87     {
  88       passiveInfo = defaultPassiveInfo;
  89       service_description = "Memory usage";
  90       use = "local-service";
  91       check_command = ["check_memory" "80" "90"];
  92     }
  93     {
  94       passiveInfo = defaultPassiveInfo;
  95       service_description = "NTP is activated and working";
  96       use = "local-service";
  97       check_command = ["check_ntp"];
  98     }
  99     {
 100       passiveInfo = defaultPassiveInfo;
 101       service_description = "No mdadm array is degraded";
 102       use = "local-service";
 103       check_command = [
 104         "check_command_output"
 105         "${mdadm}/bin/mdadm --monitor --scan -1"
 106         "^$"
 107         "-s 0 -r root"
 108       ];
 109     }
 110   ];
 111   command = {
 112     check_dns = "$USER1$/check_dns -H $ARG1$ -s $HOSTADDRESS$ $ARG2$";
 113     check_emails = "$USER2$/check_emails -H $HOSTADDRESS$ -i $USER203$ -l $ARG1$ -p $ARG2$ -s $ARG3$ -f $ARG4$";
 114     check_emails_local = "$USER2$/check_emails -H $HOSTADDRESS$ -n $ARG1$ -r $ADMINEMAIL$ -s $ARG2$ -f $ARG3$";
 115     check_backup_eriomem = "$USER2$/check_eriomem $USER208$";
 116     check_backup_eriomem_age = "$USER2$/check_backup_eriomem_age $ARG1$";
 117     check_backup_ovh_age = "$USER2$/check_backup_ovh_age $ARG1$";
 118     check_external_dns = "$USER1$/check_dns -H $ARG2$ -s $ARG1$ $ARG3$";
 119     check_ftp_database = "$USER2$/check_ftp_database";
 120     check_git = "$USER2$/check_git $USER203$";
 121     check_http = "$USER1$/check_http --sni -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -r \"$ARG3$\"";
 122     check_https = "$USER1$/check_http --sni --ssl -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -r \"$ARG3$\"";
 123     check_https_auth = "$USER1$/check_http --sni --ssl -a \"$USER202$\" -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -r \"$ARG3$\"";
 124     check_https_certificate = "$USER1$/check_http --sni --ssl -H \"$ARG1$\" -C 21,15";
 125     check_https_code = "$USER1$/check_http --sni --ssl -f stickyport -H \"$ARG1$\" -u \"$ARG2$\" -e \"$ARG3$\" -r \"$ARG4$\"";
 126     check_imap_connection = "$USER2$/check_imap_connection -u \"$USER204$\" -p \"$USER205$\" -H \"imap.immae.eu:143\"";
 127     check_local_disk = "$USER1$/check_disk -w $ARG1$ -c $ARG2$ -p $ARG3$";
 128     check_local_procs = "$USER1$/check_procs -w $ARG1$ -c $ARG2$ -s $ARG3$";
 129     check_local_load = "$USER1$/check_load -w $ARG1$ -c $ARG2$";
 130     check_local_swap = "$USER1$/check_swap -n ok -w $ARG1$ -c $ARG2$";
 131     check_local_bandwidth = "$USER2$/check_bandwidth -i=$ARG1$ -w $ARG2$ -c $ARG3$";
 132     check_memory = "$USER2$/check_mem.sh -w $ARG1$ -c $ARG2$";
 133     check_command_match = "$USER2$/check_command -c \"$ARG1$\" -C \"$ARG2$\" $ARG3$";
 134     check_command_output = "$USER2$/check_command -c \"$ARG1$\" -s 0 -o \"$ARG2$\" $ARG3$";
 135     check_command_status = "$USER2$/check_command -c \"$ARG1$\" -s \"$ARG2$\" $ARG3$";
 136     check_ntp = "$USER1$/check_ntp_time -t 30 -q -H 0.arch.pool.ntp.org";
 137     check_mailq = "$USER1$/check_mailq -s -w 1 -c 2";
 138     check_mysql_replication = "${sudo} -u mysql $USER2$/check_mysql_replication \"$ARG1$\" \"$ARG2$\"";
 139     check_postgresql_replication = "${sudo} -u postgres $USER2$/check_postgres_replication \"$ARG1$\" \"$ARG2$\" \"$ARG3$\"";
 140     check_openldap_replication = "${sudo} -u openldap $USER2$/check_openldap_replication \"$ARG1$\" \"$ARG2$\" \"$ARG3$\" \"$ARG4$\" \"$ARG5$\"";
 141     check_ovh_sms = "$USER2$/check_ovh_sms \"$USER209$\"";
 142     check_redis_replication = "${sudo} -u redis $USER2$/check_redis_replication \"$ARG1$\"";
 143     check_smtp = "$USER1$/check_smtp -H $HOSTADDRESS$ -p 25 -S -D 21,15";
 144     check_tcp = "$USER1$/check_tcp -H $HOSTADDRESS$ -p $ARG1$ -e \"$ARG2$\" -Mcrit";
 145     check_tcp_ssl = "$USER1$/check_tcp -H $HOSTADDRESS$ -p $ARG1$ -S -D 21,15";
 146
 147     check_host_alive = "$USER1$/check_ping -H $HOSTADDRESS$ -w 3000.0,80% -c 5000.0,100% -p 5";
 148     check_last_file_date = "${sudo} -u \"$ARG3$\" $USER2$/check_last_file_date \"$ARG1$\" \"$ARG2$\"";
 149     check_ok = "$USER1$/check_dummy 0 \"Dummy OK\"";
 150     check_critical = "$USER1$/check_dummy 2 \"Dummy CRITICAL\"";
 151
 152     # $OVE is to force naemon to run via shell instead of execve which fails here
 153     notify-host-by-email = "ADMINEMAIL=\"$ADMINEMAIL$\" SERVICENOTIFICATIONID=\"$SERVICENOTIFICATIONID$\" HOSTSTATE=\"$HOSTSTATE$\" HOSTOUTPUT=\"$HOSTOUTPUT$\" $USER2$/notify_by_email host \"$NOTIFICATIONTYPE$\" \"$HOSTALIAS$\" \"$LONGDATETIME$\" \"$CONTACTEMAIL$\" $OVE";
 154     # $OVE is to force naemon to run via shell instead of execve which fails here
 155     notify-service-by-email = "ADMINEMAIL=\"$ADMINEMAIL$\" SERVICENOTIFICATIONID=\"$SERVICENOTIFICATIONID$\" SERVICEDESC=\"$SERVICEDESC$\" SERVICESTATE=\"$SERVICESTATE$\" SERVICEOUTPUT=\"$SERVICEOUTPUT$\" $USER2$/notify_by_email service \"$NOTIFICATIONTYPE$\" \"$HOSTALIAS$\" \"$LONGDATETIME$\" \"$CONTACTEMAIL$\" $OVE";
 156     notify-by-slack = "HOST=\"$HOSTALIAS$\" SERVICESTATE=\"$SERVICESTATE$\" SERVICEDESC=\"$SERVICEDESC$\" SERVICEOUTPUT=\"$SERVICEOUTPUT$\" $USER2$/notify_by_slack \"$ARG1$\" \"$ARG2$\"";
 157
 158     notify-master = "$USER2$/send_nrdp.sh -u \"$USER200$\" -t \"$USER201$\" -H \"$HOSTADDRESS$\" -s \"$SERVICEDESC$\" -S \"$SERVICESTATEID$\" -o \"$SERVICEOUTPUT$ | $SERVICEPERFDATA$\"";
 159   };
 160   timeperiod = {
 161     "24x7" = {
 162       alias = "24 Hours A Day, 7 Days A Week";
 163       monday = "00:00-24:00";
 164       tuesday = "00:00-24:00";
 165       wednesday = "00:00-24:00";
 166       thursday = "00:00-24:00";
 167       friday = "00:00-24:00";
 168       saturday = "00:00-24:00";
 169       sunday = "00:00-24:00";
 170     };
 171   };
 172   servicegroup = {
 173     webstatus-webapps = { alias = "Web applications"; };
 174     webstatus-websites = { alias = "Personal websites"; };
 175     webstatus-ssl = { alias = "SSL certificates"; };
 176     webstatus-dns = { alias = "DNS resolution"; };
 177     webstatus-remote-services = { alias = "Other remote services"; };
 178     webstatus-local-services = { alias = "Other local services"; };
 179     webstatus-email = { alias = "E-mail services"; };
 180     webstatus-resources = { alias = "Local resources"; };
 181     webstatus-databases = { alias = "Databases resources"; };
 182     webstatus-backup = { alias = "Backup resources"; };
 183   };
 184   hostgroup = {
 185     webstatus-hosts = { alias = "Hosts"; };
 186   };
 187   contactgroup = {
 188     admins = { alias = "Naemon Administrators"; };
 189   };
 190   templates = {
 191     service = {
 192       generic-service = {
 193         active_checks_enabled = "1";
 194         check_freshness = "0";
 195         check_interval = "10";
 196         check_period = "24x7";
 197         contact_groups = "admins";
 198         event_handler_enabled = "1";
 199         flap_detection_enabled = "1";
 200         is_volatile = "0";
 201         max_check_attempts = "3";
 202         notification_interval = "60";
 203         notification_options = "w,u,c,r,f,s";
 204         notification_period = "24x7";
 205         notifications_enabled = if master then "1" else "0";
 206         obsess_over_service = "1";
 207         passive_checks_enabled = "1";
 208         process_perf_data = "1";
 209         retain_nonstatus_information = "1";
 210         retain_status_information = "1";
 211         retry_interval = "2";
 212       };
 213       local-service = {
 214         use = "generic-service";
 215         host_name = hostFQDN;
 216         check_interval = "5";
 217         max_check_attempts = "4";
 218         retry_interval = "1";
 219         servicegroups = "webstatus-resources";
 220       };
 221       external-service = {
 222         use = "generic-service";
 223         check_interval = "5";
 224         max_check_attempts = "4";
 225         retry_interval = "1";
 226       };
 227       web-service = {
 228         use = "generic-service";
 229         check_interval = "20";
 230         max_check_attempts = "2";
 231         retry_interval = "1";
 232       };
 233       external-web-service = {
 234         use = "generic-service";
 235         check_interval = "20";
 236         max_check_attempts = "2";
 237         retry_interval = "1";
 238       };
 239       mail-service = {
 240         use = "generic-service";
 241         check_interval = "15";
 242         max_check_attempts = "1";
 243         retry_interval = "1";
 244       };
 245       dns-service = {
 246         use = "generic-service";
 247         check_interval = "120";
 248         notification_interval = "120";
 249         max_check_attempts = "5";
 250         retry_interval = "5";
 251       };
 252     };
 253     # No contact, we go through master
 254     contact = {
 255       generic-contact = {
 256         host_notification_commands = "notify-host-by-email";
 257         host_notification_options = "d,u,r,f,s";
 258         host_notification_period = "24x7";
 259         service_notification_commands = "notify-service-by-email";
 260         service_notification_options = "w,u,c,r,f,s";
 261         service_notification_period = "24x7";
 262       };
 263     };
 264     host = {
 265       generic-host = {
 266         event_handler_enabled = "1";
 267         flap_detection_enabled = "1";
 268         notification_period = "24x7";
 269         notifications_enabled = "1";
 270         process_perf_data = "1";
 271         retain_nonstatus_information = "1";
 272         retain_status_information = "1";
 273       };
 274       linux-server = {
 275         check_command = "check_host_alive";
 276         check_interval = "5";
 277         check_period = "24x7";
 278         contact_groups = "admins";
 279         max_check_attempts = "10";
 280         notification_interval = "120";
 281         notification_options = "d,u,r,f";
 282         retry_interval = "1";
 283       };
 284     };
 285   };
 286 }