feat: adds alerts for thin pool usage

config/prometheus/prometheus_rules.yaml

@@ -24,3 +24,41 @@
       "for": "5m"
       "labels":
         "severity": "critical"
+  - "name": "thin-pool-alert.rules"
+    "rules":
+    - "alert": "ThinPoolDataUsageAtThresholdNearFull"
+      "annotations":
+        "description": "Thin pool in the VolumeGroup is nearing full. Data deletion or thin pool expansion is required."
+        "message": "Thin Pool data utilization in the VolumeGroup {{ $labels.device_class }} has crossed 75 % on node {{ $labels.node }}. Free up some space or expand the thin pool."
+      "expr": |
+        topolvm_thinpool_data_percent > 75.00  and topolvm_thinpool_data_percent < 85.00
+      "for": "5m"
+      "labels":
+        "severity": "warning"
+    - "alert": "ThinPoolDataUsageAtThresholdCritical"
+      "annotations":
+        "description": "Thin pool in the VolumeGroup is critically full. Data deletion or thin pool expansion is required."
+        "message": "Thin Pool data utilization in the VolumeGroup {{ $labels.device_class }} has crossed 85 % on node {{ $labels.node }}. Free up some space or expand the thin pool immediately."
+      "expr": |
+        topolvm_thinpool_data_percent > 85.00
+      "for": "5m"
+      "labels":
+        "severity": "critical"
+    - "alert": "ThinPoolMetaDataUsageAtThresholdNearFull"
+      "annotations":
+        "description": "Thin pool metadata utitlization in the VolumeGroup is nearing full. Data deletion or thin pool expansion is required."
+        "message": "Thin Pool metadata utilization in the VolumeGroup {{ $labels.device_class }} has crossed 75 % on node {{ $labels.node }}. Free up some space or expand the thin pool."
+      "expr": |
+        topolvm_thinpool_metadata_percent > 75.00  and topolvm_thinpool_data_percent < 85.00
+      "for": "5m"
+      "labels":
+        "severity": "warning"
+    - "alert": "ThinPoolMetaDataUsageAtThresholdCritical"
+      "annotations":
+        "description": "Thin pool metadata ultilization in the VolumeGroup is critically full. Data deletion or thin pool expansion is required."
+        "message": "Thin Pool metadata utilization in the VolumeGroup {{ $labels.device_class }} has crossed 85 % on node {{ $labels.node }}. Free up some space or expand the thin pool immediately."
+      "expr": |
+        topolvm_thinpool_metadata_percent > 85.00
+      "for": "5m"
+      "labels":
+        "severity": "critical"

monitoring/alerts/vgalerts.libsonnet

@@ -34,6 +34,67 @@
           },
 	],
       },
+      {
+        name: 'thin-pool-alert.rules',
+        rules: [
+          {
+            alert: 'ThinPoolDataUsageAtThresholdNearFull',
+            expr: |||
+              topolvm_thinpool_data_percent > %(thinPoolUsageThresholdNearFull)0.2f  and topolvm_thinpool_data_percent < %(thinPoolUsageThresholdCritical)0.2f
+            ||| % $._config,
+            'for': $._config.thinPoolUsageThresholdAlertTime,
+            labels: {
+              severity: 'warning',
+            },
+            annotations: {
+              description: 'Thin pool in the VolumeGroup is nearing full. Data deletion or thin pool expansion is required.',
+              message: 'Thin Pool data utilization in the VolumeGroup {{ $labels.device_class }} has crossed %.0f %% on node {{ $labels.node }}. Free up some space or expand the thin pool.' % ($._config.thinPoolUsageThresholdNearFull),
+            },
+          },
+          {
+            alert: 'ThinPoolDataUsageAtThresholdCritical',
+            expr: |||
+              topolvm_thinpool_data_percent > %(thinPoolUsageThresholdCritical)0.2f
+            ||| % $._config,
+            'for': $._config.thinPoolUsageThresholdAlertTime,
+            labels: {
+              severity: 'critical',
+            },
+            annotations: {
+              description: 'Thin pool in the VolumeGroup is critically full. Data deletion or thin pool expansion is required.',
+              message: 'Thin Pool data utilization in the VolumeGroup {{ $labels.device_class }} has crossed %.0f %% on node {{ $labels.node }}. Free up some space or expand the thin pool immediately.' % ($._config.thinPoolUsageThresholdCritical),
+            },
+          },
+          {
+            alert: 'ThinPoolMetaDataUsageAtThresholdNearFull',
+            expr: |||
+              topolvm_thinpool_metadata_percent > %(thinPoolUsageThresholdNearFull)0.2f  and topolvm_thinpool_data_percent < %(thinPoolUsageThresholdCritical)0.2f
+            ||| % $._config,
+            'for': $._config.thinPoolUsageThresholdAlertTime,
+            labels: {
+              severity: 'warning',
+            },
+            annotations: {
+              description: 'Thin pool metadata utitlization in the VolumeGroup is nearing full. Data deletion or thin pool expansion is required.',
+              message: 'Thin Pool metadata utilization in the VolumeGroup {{ $labels.device_class }} has crossed %.0f %% on node {{ $labels.node }}. Free up some space or expand the thin pool.' % ($._config.thinPoolUsageThresholdNearFull),
+            },
+          },
+          {
+            alert: 'ThinPoolMetaDataUsageAtThresholdCritical',
+            expr: |||
+              topolvm_thinpool_metadata_percent > %(thinPoolUsageThresholdCritical)0.2f
+            ||| % $._config,
+            'for': $._config.thinPoolUsageThresholdAlertTime,
+            labels: {
+              severity: 'critical',
+            },
+            annotations: {
+              description: 'Thin pool metadata ultilization in the VolumeGroup is critically full. Data deletion or thin pool expansion is required.',
+              message: 'Thin Pool metadata utilization in the VolumeGroup {{ $labels.device_class }} has crossed %.0f %% on node {{ $labels.node }}. Free up some space or expand the thin pool immediately.' % ($._config.thinPoolUsageThresholdCritical),
+            },
+          },
+	],
+      },
     ],
   },
 }

monitoring/config.libsonnet

@@ -8,6 +8,14 @@
 
     // alert durations
     volumegroupUsageThresholdAlertTime: '5m',
+
+    // thin pool data and metadata usage percentage threshold near full
+    thinPoolUsageThresholdNearFull : 75,
+
+    // thin pool data and metadata usage percentage threshold critical
+    thinPoolUsageThresholdCritical : 85,
+
+    // alert durations
+    thinPoolUsageThresholdAlertTime: '5m',
   },
 }
-