systems/dilion/monitoring.nix

   1 { lib, pkgs, config, name, monitoring, ... }:
   2 let
   3   hostFQDN = config.hostEnv.fqdn;
   4   zfs_snapshot = name: {
   5     service_description = "ZFS snapshot ${name} happened not too long ago";
   6     use = "local-service";
   7     check_command = ["check_zfs_snapshot" name];
   8     __passive_servicegroups = "webstatus-resources";
   9   };
  10 in
  11 {
  12   config.myServices.monitoring.activatedPlugins = [ "memory" "command" "bandwidth" "mdadm" "zfs" "notify-secondary"];
  13   config.myServices.monitoring.objects = lib.mkMerge [
  14     (monitoring.lib.objectsCommon {
  15       inherit hostFQDN;
  16       hostName = name;
  17       master = false;
  18       processWarn = "250"; processAlert = "400";
  19       loadWarn = "1.0"; loadAlert = "1.2";
  20       interface = builtins.head (builtins.attrNames config.networking.interfaces);
  21     })
  22
  23     {
  24       service = [
  25         {
  26           service_description = "No mdadm array is degraded";
  27           use = "local-service";
  28           check_command = ["check_mdadm"];
  29           __passive_servicegroups = "webstatus-resources";
  30         }
  31         {
  32           service_description = "No ZFS pool is degraded";
  33           use = "local-service";
  34           check_command = ["check_zfs"];
  35           __passive_servicegroups = "webstatus-resources";
  36         }
  37         (zfs_snapshot "zpool/backup/eldiron/zpool/root")
  38         (zfs_snapshot "zpool/backup/eldiron/zpool/root/etc")
  39         (zfs_snapshot "zpool/backup/eldiron/zpool/root/var")
  40       ];
  41     }
  42   ];
  43 }