From 59dc6dca5c9bb366033090518ed2831b8acd422c Mon Sep 17 00:00:00 2001
From: samber <samber@users.noreply.github.com>
Date: Sat, 24 Feb 2024 19:15:25 +0000
Subject: [PATCH] Publish

---
 .../smartctl-exporter.yml                     | 56 ++++++++++++-------
 1 file changed, 37 insertions(+), 19 deletions(-)

diff --git a/dist/rules/s.m.a.r.t-device-monitoring/smartctl-exporter.yml b/dist/rules/s.m.a.r.t-device-monitoring/smartctl-exporter.yml
index 1946c38..4334bf0 100644
--- a/dist/rules/s.m.a.r.t-device-monitoring/smartctl-exporter.yml
+++ b/dist/rules/s.m.a.r.t-device-monitoring/smartctl-exporter.yml
@@ -5,46 +5,64 @@ groups:
   rules:
 
     - alert: SmartDeviceTemperatureWarning
-      expr: 'smartctl_device_temperature > 60'
-      for: 2m
+      expr: 'avg_over_time(smartctl_device_temperature{temperature_type="current"} [10m]) > 60'
+      for: 0m
       labels:
         severity: warning
       annotations:
-        summary: Smart device temperature warning (instance {{ $labels.instance }})
-        description: "Device temperature  warning (instance {{ $labels.instance }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
+        summary: SMART device temperature warning (instance {{ $labels.instance }})
+        description: "Device temperature warning (instance {{ $labels.instance }}, drive {{ $labels.device }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
 
     - alert: SmartDeviceTemperatureCritical
-      expr: 'smartctl_device_temperature > 80'
-      for: 2m
+      expr: 'max_over_time(smartctl_device_temperature{temperature_type="current"} [10m]) >= 70'
+      for: 0m
       labels:
         severity: critical
       annotations:
-        summary: Smart device temperature critical (instance {{ $labels.instance }})
-        description: "Device temperature critical  (instance {{ $labels.instance }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
+        summary: SMART device temperature critical (instance {{ $labels.instance }})
+        description: "Device temperature critical (instance {{ $labels.instance }}, drive {{ $labels.device }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
+
+    - alert: SmartDeviceTemperatureWasOverTripValue
+      expr: 'max_over_time(smartctl_device_temperature{temperature_type="current"} [10m]) >= on(device, instance) smartctl_device_temperature{temperature_type="drive_trip"}'
+      for: 0m
+      labels:
+        severity: critical
+      annotations:
+        summary: SMART device temperature was over trip value (instance {{ $labels.instance }})
+        description: "Device temperature over trip value (instance {{ $labels.instance }}, drive {{ $labels.device }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
+
+    - alert: SmartStatus
+      expr: 'smartctl_device_smart_status != 1'
+      for: 0m
+      labels:
+        severity: critical
+      annotations:
+        summary: SMART status (instance {{ $labels.instance }})
+        description: "Device has a SMART status failure (instance {{ $labels.instance }}, drive {{ $labels.device }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
 
     - alert: SmartCriticalWarning
       expr: 'smartctl_device_critical_warning > 0'
-      for: 15m
+      for: 0m
       labels:
         severity: critical
       annotations:
-        summary: Smart critical warning (instance {{ $labels.instance }})
-        description: "device has critical warning (instance {{ $labels.instance }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
+        summary: SMART critical warning (instance {{ $labels.instance }})
+        description: "Disk controller has critical warning (instance {{ $labels.instance }}, drive {{ $labels.device }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
 
     - alert: SmartMediaErrors
       expr: 'smartctl_device_media_errors > 0'
-      for: 15m
+      for: 0m
       labels:
         severity: critical
       annotations:
-        summary: Smart media errors (instance {{ $labels.instance }})
-        description: "device has media errors (instance {{ $labels.instance }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
+        summary: SMART media errors (instance {{ $labels.instance }})
+        description: "Disk controller detected media errors (instance {{ $labels.instance }}, drive {{ $labels.device }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
 
-    - alert: SmartNvmeWearoutIndicator
-      expr: 'smartctl_device_available_spare{device=~"nvme.*"} < smartctl_device_available_spare_threshold{device=~"nvme.*"}'
-      for: 15m
+    - alert: SmartWearoutIndicator
+      expr: 'smartctl_device_available_spare < smartctl_device_available_spare_threshold'
+      for: 0m
       labels:
         severity: critical
       annotations:
-        summary: Smart NVME Wearout Indicator (instance {{ $labels.instance }})
-        description: "NVMe device is wearing out (instance {{ $labels.instance }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
+        summary: SMART Wearout Indicator (instance {{ $labels.instance }})
+        description: "Device is wearing out (instance {{ $labels.instance }}, drive {{ $labels.device }})\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"