manatee: Update monitoring in home-assistant

- Add all disks to smartd - Generate home-assistant config in nix - Add metrics for all HDDs
2026-05-30 16:54:40 +00:00
parent d291633fe2
commit 4a63c4eb5e
4 changed files with 353 additions and 141 deletions
@@ -0,0 +1,159 @@
+{
+  pkgs,
+  lib,
+  config,
+  ...
+}:
+let
+  enabled = config.mod.disk-smart.enable;
+
+  disks = [
+    { path = "/dev/disk/by-id/ata-ST8000VN004-3CP101_WWZ8QCG4"; name = "seagate_8tb_1"; label = "Seagate 8TB #1"; }
+    { path = "/dev/disk/by-id/ata-ST8000VN004-3CP101_WWZ8QDJ5"; name = "seagate_8tb_2"; label = "Seagate 8TB #2"; }
+    { path = "/dev/disk/by-id/ata-TOSHIBA_MG10ACA20TE_85K2A0UCF4MJ"; name = "toshiba_20tb_1"; label = "Toshiba 20TB #1"; }
+    { path = "/dev/disk/by-id/ata-TOSHIBA_MG10ACA20TE_85K2A0V6F4MJ"; name = "toshiba_20tb_2"; label = "Toshiba 20TB #2"; }
+  ];
+
+  outputDir = "/var/lib/disk-smart";
+
+  collectScript = pkgs.writeShellScript "disk-smart-collect" ''
+    set -euo pipefail
+    export PATH="${lib.makeBinPath [ pkgs.smartmontools pkgs.jq pkgs.coreutils ]}"
+
+    mkdir -p ${outputDir}
+
+    result="{"
+
+    ${lib.concatMapStringsSep "\n" (disk: ''
+      raw=$(smartctl -j -A -H ${disk.path} 2>/dev/null || true)
+
+      temp=$(echo "$raw" | jq -r '.temperature.current // empty')
+      power_on=$(echo "$raw" | jq -r '.power_on_time.hours // empty')
+      smart_status=$(echo "$raw" | jq -r '.smart_status.passed // empty')
+      reallocated=$(echo "$raw" | jq -r '[.ata_smart_attributes.table[] | select(.name == "Reallocated_Sector_Ct")][0].raw.value // empty')
+      pending=$(echo "$raw" | jq -r '[.ata_smart_attributes.table[] | select(.name == "Current_Pending_Sector")][0].raw.value // empty')
+
+      result="$result\"${disk.name}\":{\"temperature\":$temp,\"power_on_hours\":$power_on,\"smart_passed\":$smart_status,\"reallocated_sectors\":$reallocated,\"pending_sectors\":$pending},"
+    '') disks}
+
+    # Remove trailing comma, close object
+    result="''${result%,}}"
+
+    echo "$result" | jq . > ${outputDir}/smart.json.tmp
+    mv ${outputDir}/smart.json.tmp ${outputDir}/smart.json
+  '';
+
+  indent = prefix: s:
+    lib.concatMapStringsSep "\n"
+      (line: if line == "" then line else prefix + line)
+      (lib.splitString "\n" s);
+
+  mkSensor = disk: ''
+- name: "${disk.label} Temperature"
+  value_template: "{{ value_json.${disk.name}.temperature }}"
+  unit_of_measurement: "°C"
+  device_class: temperature
+  state_class: measurement
+- name: "${disk.label} Power On Hours"
+  value_template: "{{ value_json.${disk.name}.power_on_hours }}"
+  unit_of_measurement: "h"
+  state_class: total_increasing
+- name: "${disk.label} SMART Passed"
+  value_template: "{{ value_json.${disk.name}.smart_passed }}"
+- name: "${disk.label} Reallocated Sectors"
+  value_template: "{{ value_json.${disk.name}.reallocated_sectors }}"
+  state_class: measurement
+- name: "${disk.label} Pending Sectors"
+  value_template: "{{ value_json.${disk.name}.pending_sectors }}"
+  state_class: measurement
+'';
+
+  sensorYaml = indent "      " (lib.concatMapStrings mkSensor disks);
+
+  sectorEntities = lib.concatMap (disk: [
+    "sensor.${disk.name}_reallocated_sectors"
+    "sensor.${disk.name}_pending_sectors"
+  ]) disks;
+
+  sectorEntitiesYaml = lib.concatMapStringsSep "\n"
+    (id: "          - ${id}") sectorEntities;
+
+  smartPassedEntities = map (disk: "sensor.${disk.name}_smart_passed") disks;
+
+  smartPassedEntitiesYaml = lib.concatMapStringsSep "\n"
+    (id: "          - ${id}") smartPassedEntities;
+in
+{
+  options = {
+    mod.disk-smart = {
+      enable = lib.mkEnableOption "Enable disk SMART monitoring module";
+    };
+  };
+
+  config = lib.mkIf enabled {
+    mod.home-assistant.extraConfig = ''
+rest:
+  - resource: http://127.0.0.1:9633/smart.json
+    scan_interval: 60
+    sensor:
+${sensorYaml}
+automation disk_smart:
+  - alias: "Disk sector count increased"
+    trigger:
+      - platform: state
+        entity_id:
+${sectorEntitiesYaml}
+    condition:
+      - condition: template
+        value_template: "{{ trigger.from_state.state | int(-1) >= 0 and trigger.to_state.state | int(0) > trigger.from_state.state | int(0) }}"
+    action:
+      - service: notify.mobile_app_pixel_9_pro
+        data:
+          title: "Disk SMART warning"
+          message: "{{ trigger.to_state.attributes.friendly_name }} increased from {{ trigger.from_state.state }} to {{ trigger.to_state.state }}"
+  - alias: "Disk SMART check failed"
+    trigger:
+      - platform: state
+        entity_id:
+${smartPassedEntitiesYaml}
+    condition:
+      - condition: template
+        value_template: "{{ trigger.to_state.state | lower == 'false' }}"
+    action:
+      - service: notify.mobile_app_pixel_9_pro
+        data:
+          title: "Disk SMART FAILURE"
+          message: "{{ trigger.to_state.attributes.friendly_name }} reports SMART failure — drive is likely failing"
+'';
+
+    systemd.services.disk-smart-collect = {
+      description = "Collect disk SMART data";
+      serviceConfig = {
+        Type = "oneshot";
+        ExecStart = collectScript;
+      };
+    };
+
+    systemd.timers.disk-smart-collect = {
+      description = "Periodically collect disk SMART data";
+      wantedBy = [ "timers.target" ];
+      timerConfig = {
+        OnBootSec = "1min";
+        OnUnitActiveSec = "1min";
+      };
+    };
+
+    services.nginx.virtualHosts."127.0.0.1" = {
+      listen = [
+        { addr = "127.0.0.1"; port = 9633; }
+      ];
+
+      locations."= /smart.json" = {
+        alias = "${outputDir}/smart.json";
+        extraConfig = ''
+          default_type application/json;
+        '';
+      };
+    };
+  };
+}