Alert for RAID batteries that have been recharging for too long

diff --git a/cookbooks/prometheus/templates/default/alert_rules.yml.erb b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
index 8f2986e6ab85340d339be8e23f9adc219f0911ac..9a3c52920bd3c7793ffb88e1b21ea615368cee5b 100644 (file)
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -558,6 +558,11 @@ groups:
         for: 5m
         labels:
           alertgroup: "{{ $labels.instance }}"
         for: 5m
         labels:
           alertgroup: "{{ $labels.instance }}"

+      - alert: raid controller battery recharging
+        expr: ohai_controller_info{battery_status="recharging"} > 0
+        for: 4h
+        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