]> git.openstreetmap.org Git - chef.git/commitdiff
Add an alert for RAID controller battery failures
authorTom Hughes <tom@compton.nu>
Fri, 24 Feb 2023 14:28:18 +0000 (14:28 +0000)
committerTom Hughes <tom@compton.nu>
Fri, 24 Feb 2023 14:28:18 +0000 (14:28 +0000)
cookbooks/prometheus/templates/default/alert_rules.yml.erb

index 1736da60cf16138b80cf5d8cc50ee60e68b94555..01d8ae00f654fce5a0120e3427928fd340c4d8df 100644 (file)
@@ -530,6 +530,11 @@ groups:
           alertgroup: "prometheus"
   - name: raid
     rules:
           alertgroup: "prometheus"
   - name: raid
     rules:
+      - alert: raid controller battery failed
+        expr: ohai_controller_info{battery_status="failed"} > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
       - alert: raid array degraded
         expr: ohai_array_info{status="degraded"} > 0
         for: 5m
       - alert: raid array degraded
         expr: ohai_array_info{status="degraded"} > 0
         for: 5m