Add monitoring script with smartctl

7 files changed, 71 insertions, 7 deletions

diff --git a/systems/backup-2/flake.lock b/systems/backup-2/flake.lock
index a8d55bf..d6e56a5 100644
--- a/systems/backup-2/flake.lock
+++ b/systems/backup-2/flake.lock
@@ -389,7 +389,7 @@
       },
       "locked": {
         "lastModified": 1,
-        "narHash": "sha256-PicwNqwQaCKtN5zPpTfpSDRXbyn9xh8aHqoPnm9cFz8=",
+        "narHash": "sha256-A50vXg0qfzUhyQZKToEseYsLgtaYke41KE/4uuyj+PY=",
         "path": "../../flakes/private/monitoring",
         "type": "path"
       },
diff --git a/systems/dilion/flake.lock b/systems/dilion/flake.lock
index 3096384..a690504 100644
--- a/systems/dilion/flake.lock
+++ b/systems/dilion/flake.lock
@@ -207,7 +207,7 @@
       },
       "locked": {
         "lastModified": 1,
-        "narHash": "sha256-PicwNqwQaCKtN5zPpTfpSDRXbyn9xh8aHqoPnm9cFz8=",
+        "narHash": "sha256-A50vXg0qfzUhyQZKToEseYsLgtaYke41KE/4uuyj+PY=",
         "path": "../../flakes/private/monitoring",
         "type": "path"
       },
diff --git a/systems/dilion/monitoring.nix b/systems/dilion/monitoring.nix
index af7f224..5c4e5b0 100644
--- a/systems/dilion/monitoring.nix
+++ b/systems/dilion/monitoring.nix
@@ -9,7 +9,7 @@ let
   };
 in
 {
-  config.myServices.monitoring.activatedPlugins = [ "memory" "command" "bandwidth" "mdadm" "zfs" "notify-secondary"];
+  config.myServices.monitoring.activatedPlugins = [ "memory" "command" "bandwidth" "mdadm" "zfs" "notify-secondary" "smartctl" ];
   config.myServices.monitoring.objects = lib.mkMerge [
     (monitoring.lib.objectsCommon {
       inherit hostFQDN;
@@ -23,6 +23,38 @@ in
     {
       service = [
         {
+          service_description = "Disk /dev/disk/by-id/ata-Micron_1100_MTFDDAK512TBN_171216BD2A4B is sane";
+          use = "local-service";
+          check_command = [ "check_smartctl" "/dev/disk/by-id/ata-Micron_1100_MTFDDAK512TBN_171216BD2A4B" ];
+          __passive_servicegroups = "webstatus-resources";
+
+          check_interval = 60;
+        }
+        {
+          service_description = "Disk /dev/disk/by-id/ata-Micron_1100_MTFDDAK512TBN_171216BD2A70 is sane";
+          use = "local-service";
+          check_command = [ "check_smartctl" "/dev/disk/by-id/ata-Micron_1100_MTFDDAK512TBN_171216BD2A70" ];
+          __passive_servicegroups = "webstatus-resources";
+
+          check_interval = 60;
+        }
+        {
+          service_description = "Disk /dev/disk/by-id/ata-TOSHIBA_MG04ACA400N_96K1K87YFVLC is sane";
+          use = "local-service";
+          check_command = [ "check_smartctl" "/dev/disk/by-id/ata-TOSHIBA_MG04ACA400N_96K1K87YFVLC" ];
+          __passive_servicegroups = "webstatus-resources";
+
+          check_interval = 60;
+        }
+        {
+          service_description = "Disk /dev/disk/by-id/ata-WDC_WD3000FYYZ-01UL1B2_WD-WMC1F0E4X6WP is sane";
+          use = "local-service";
+          check_command = [ "check_smartctl" "/dev/disk/by-id/ata-WDC_WD3000FYYZ-01UL1B2_WD-WMC1F0E4X6WP" ];
+          __passive_servicegroups = "webstatus-resources";
+
+          check_interval = 60;
+        }
+        {
           service_description = "No mdadm array is degraded";
           use = "local-service";
           check_command = ["check_mdadm"];
diff --git a/systems/eldiron/flake.lock b/systems/eldiron/flake.lock
index f5b9184..461bebb 100644
--- a/systems/eldiron/flake.lock
+++ b/systems/eldiron/flake.lock
@@ -2038,7 +2038,7 @@
       },
       "locked": {
         "lastModified": 1,
-        "narHash": "sha256-PicwNqwQaCKtN5zPpTfpSDRXbyn9xh8aHqoPnm9cFz8=",
+        "narHash": "sha256-A50vXg0qfzUhyQZKToEseYsLgtaYke41KE/4uuyj+PY=",
         "path": "../../flakes/private/monitoring",
         "type": "path"
       },
diff --git a/systems/eldiron/monitoring.nix b/systems/eldiron/monitoring.nix
index 2aa7f8f..92da7cf 100644
--- a/systems/eldiron/monitoring.nix
+++ b/systems/eldiron/monitoring.nix
@@ -4,7 +4,7 @@ let
   emailCheck = monitoring.lib.emailCheck config.myEnv.monitoring.email_check;
 in
 {
-  config.myServices.monitoring.activatedPlugins = [ "memory" "command" "bandwidth" "emails" "mdadm" "postfix" "postgresql" "zfs" "notify-secondary"];
+  config.myServices.monitoring.activatedPlugins = [ "memory" "command" "bandwidth" "emails" "mdadm" "postfix" "postgresql" "zfs" "notify-secondary" "smartctl" ];
   config.myServices.monitoring.pluginsArgs.postgresql.package = config.myServices.databases.postgresql.package;
   config.myServices.monitoring.objects = lib.mkMerge [
     (monitoring.lib.objectsCommon {
@@ -19,6 +19,38 @@ in
     {
       service = [
         {
+          service_description = "Disk /dev/disk/by-id/ata-ST33000650NS_Z29540DM is sane";
+          use = "local-service";
+          check_command = [ "check_smartctl" "/dev/disk/by-id/ata-ST33000650NS_Z29540DM" ];
+          __passive_servicegroups = "webstatus-resources";
+
+          check_interval = 60;
+        }
+        {
+          service_description = "Disk /dev/disk/by-id/ata-ST33000650NS_Z296JGJ4 is sane";
+          use = "local-service";
+          check_command = [ "check_smartctl" "/dev/disk/by-id/ata-ST33000650NS_Z296JGJ4" ];
+          __passive_servicegroups = "webstatus-resources";
+
+          check_interval = 60;
+        }
+        {
+          service_description = "Disk /dev/disk/by-id/ata-SAMSUNG_MZ7LM480HMHQ-00005_S2UJNX0HA05895 is sane";
+          use = "local-service";
+          check_command = [ "check_smartctl" "/dev/disk/by-id/ata-SAMSUNG_MZ7LM480HMHQ-00005_S2UJNX0HA05895" ];
+          __passive_servicegroups = "webstatus-resources";
+
+          check_interval = 60;
+        }
+        {
+          service_description = "Disk /dev/disk/by-id/ata-Micron_5200_MTFDDAK480TDC_18011BEE03B3 is sane";
+          use = "local-service";
+          check_command = [ "check_smartctl" "/dev/disk/by-id/ata-Micron_5200_MTFDDAK480TDC_18011BEE03B3" ];
+          __passive_servicegroups = "webstatus-resources";
+
+          check_interval = 60;
+        }
+        {
           service_description = "No mdadm array is degraded";
           use = "local-service";
           check_command = ["check_mdadm"];
diff --git a/systems/monitoring-1/flake.lock b/systems/monitoring-1/flake.lock
index ed711f4..643683a 100644
--- a/systems/monitoring-1/flake.lock
+++ b/systems/monitoring-1/flake.lock
@@ -277,7 +277,7 @@
       },
       "locked": {
         "lastModified": 1,
-        "narHash": "sha256-PicwNqwQaCKtN5zPpTfpSDRXbyn9xh8aHqoPnm9cFz8=",
+        "narHash": "sha256-A50vXg0qfzUhyQZKToEseYsLgtaYke41KE/4uuyj+PY=",
         "path": "../../flakes/private/monitoring",
         "type": "path"
       },
diff --git a/systems/quatresaisons/flake.lock b/systems/quatresaisons/flake.lock
index a76103c..afb007b 100644
--- a/systems/quatresaisons/flake.lock
+++ b/systems/quatresaisons/flake.lock
@@ -239,7 +239,7 @@
       },
       "locked": {
         "lastModified": 1,
-        "narHash": "sha256-PicwNqwQaCKtN5zPpTfpSDRXbyn9xh8aHqoPnm9cFz8=",
+        "narHash": "sha256-A50vXg0qfzUhyQZKToEseYsLgtaYke41KE/4uuyj+PY=",
         "path": "../../flakes/private/monitoring",
         "type": "path"
       },