From: Tom Hughes <tom@compton.nu>
Date: Mon, 25 Jul 2022 11:52:28 +0000 (+0100)
Subject: Add alerts for degraded raid arrays and failed raid disks
X-Git-Url: https://git.openstreetmap.org./chef.git/commitdiff_plain/e442c6a6365b6be91e350d9fe79d3a6227e7295c?ds=inline

Add alerts for degraded raid arrays and failed raid disks
---

diff --git a/cookbooks/prometheus/templates/default/alert_rules.yml.erb b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
index c108451ab..23bb157e4 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -422,6 +422,18 @@ groups:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           queries: "{{ $value }}"
+  - name: raid
+    rules:
+      - alert: raid array degraded
+        expr: ohai_array_info{status="degraded"} > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
+      - alert: raid disk failed
+        expr: ohai_disk_info{status="failed"} > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
   - name: smart
     rules:
       - alert: smart failure