add smartmon_sata failure as a mixin in kube-prometheus

build/kube-prometheus/README.md

@@ -162,11 +162,7 @@ available in https://github.com/prometheus/alertmanager/blob/main/doc/alertmanag
   }
   ```
 
-- To add a custom prometheus rule as a mixin, create a mixin.libsonnet file in the relevant folder under the `mixins` folder and generate the `prometheus.yaml` for it. One way to do generate a YAML file from a .libsonnet file is using a jsonnet command similar to this:
-
-```
-jsonnet -e '(import "mixin.libsonnet").prometheusAlerts' | gojsontoyaml > prometheus.yaml
-```
+- To add a custom prometheus rule as a mixin, create a mixin.libsonnet file in the relevant folder under the `mixins` folder.
 
 - In the case when your mixin is supposed to trigger a Prometheus alert and <b>all you want is to test</b> whether it works, do this:
 

build/kube-prometheus/common-template.jsonnet

@@ -85,6 +85,7 @@ local default_vars = {
     'velero',
     'whoami',
     'zfs-localpv',
+    'smartmon',
   ],
   prometheus_operator_resources: {
     limits: { memory: '80Mi' },
@@ -169,6 +170,7 @@ local default_vars = {
     'argo-cd-sync-state': true,
     rabbitmq: false,
     'monitor-prometheus-stack': false,
+    smartmon: false,
   },
   mixin_configs: {
     // Example:
@@ -245,6 +247,11 @@ local mixins = remove_nulls([
     (import 'mixins/monitoring/mixin.libsonnet'),
     vars,
   ),
+  addMixin(
+    'smartmon',
+    (import 'mixins/smartmon/mixin.libsonnet'),
+    vars,
+  ),
 ]);
 
 local scrape_namespaces = std.uniq(std.sort(std.flattenArrays(

build/kube-prometheus/mixins/smartmon/mixin.libsonnet

@@ -0,0 +1,29 @@
+{
+  _config+:: {
+    selector: '',
+  },
+
+  prometheusAlerts+:: {
+    groups+: [
+      {
+        name: 'SmartMon',
+        rules: [
+          {
+            alert: 'SmartMonUdmaCrcErrorCountRawValue',
+            expr: 'sum by (instance, device) (smartctl_device_attribute{attribute_name="UDMA_CRC_Error_Count", attribute_value_type="raw"} >= smartctl_device_attribute{attribute_name="UDMA_CRC_Error_Count", attribute_value_type="worst"})',
+            'for': '3h',
+            labels: {
+              severity: 'critical',
+              alert_id: 'SmartMonUdmaCrcErrorCountRawValue',
+            },
+            annotations: {
+              description: 'Disk **{{ .Labels.device }}**  has disk sata failure on instance **{{ .Labels.instance }}**
+          UDMA_CRC_Error_Count - The number of errors related to data transfer over the interface. A value of **{{ .Value }}** is concerning and indicates potential issues with the data cable or connections.',
+              summary: 'The device **{{ .Labels.device }}** has disk sata failures.'
+            },
+          },
+        ],
+      },
+    ],
+  },
+}