From: Tom Hughes <tom@compton.nu>
Date: Fri, 24 Feb 2023 14:28:18 +0000 (+0000)
Subject: Add an alert for RAID controller battery failures
X-Git-Url: https://git.openstreetmap.org./chef.git/commitdiff_plain/4a1a4f51747e0919a67e081b03ae6f853669c9c0

Add an alert for RAID controller battery failures
---

diff --git a/cookbooks/prometheus/templates/default/alert_rules.yml.erb b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
index 1736da60c..01d8ae00f 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -530,6 +530,11 @@ groups:
           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