ajout surveillance /var

tmpl/alert-rules.yml

@@ -34,8 +34,27 @@ groups:
   - alert: filesystem_threshold_exceeded
     expr: node_filesystem_avail{job="%%{job_name_node}",mountpoint="/"} / node_filesystem_size{job="%%{job_name_node}"}
       * 100 < 20
+    for: 2m
     annotations:
       description: This device's filesystem usage has exceeded the threshold with
         a value of {{ $value }}.
       summary: Instance {{ $labels.instance }} filesystem usage is dangerously high
 
+  # Heavy "/var" use
+  - alert: var_filesystem_threshold_exceeded
+    expr: node_filesystem_avail{job="node",mountpoint="/var"} / node_filesystem_size{job="node"}
+      * 100 < 20
+    annotations:
+      description: This device's filesystem usage has exceeded the threshold with
+        a value of {{ $value }}.
+      summary: Instance {{ $labels.instance }} filesystem usage is dangerously high
+  
+  # Heavy CPU temperature
+  - alert: cpu_temp_threshold_exceeded
+    expr: avg(node_hwmon_temp_celsius{job="node"}) BY (instance)
+      > 50
+    annotations:
+      description: This device's cpu temperature has exceeded the threshold with a value
+        of {{ $value }}.
+      summary: Instance {{ $labels.instance }} CPU temperature is dangerously high
+

tmpl/alertmanager.yml

@@ -81,6 +81,7 @@ route:
     - match:
         %%{sroute.alSubRouteMatchSource}: %%alSubRouteMatchValue
       receiver: %%alSubRouteMatchReceiver
+      continue: true
       %end if
     %end for
   %end if
@@ -89,6 +90,7 @@ route:
   - match:
       %%{rt.alRouteMatchSource}: %%{rt.alRouteMatchValue}
     receiver: %%rt.alRouteMatchReceiver
+    continue: true
 
   %if not %%is_empty('alSubRoute')
     routes:
@@ -97,6 +99,7 @@ route:
     - match:
         %%{sroute.alSubRouteMatchSource}: %%{sroute.alSubRouteMatchValue}
       receiver: %%sroute.alSubRouteMatchReceiver
+      continue: true
       %end if
     %end for
   %end if