From 5bbe6932cd6068a287a109cf811c075ab7d3bb47 Mon Sep 17 00:00:00 2001
From: Tom Hughes <tom@compton.nu>
Date: Sun, 28 Aug 2022 22:25:54 +0100
Subject: [PATCH] Add alerts for rasdaemon events

---
 .../templates/default/alert_rules.yml.erb        | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/cookbooks/prometheus/templates/default/alert_rules.yml.erb b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
index ab8cfe3ec..2920d2b16 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -444,6 +444,22 @@ groups:
         for: 5m
         labels:
           alertgroup: "{{ $labels.instance }}"
+  - name: rasdaemon
+    rules:
+      - alert: memory controller errors
+        expr: increase(rasdaemon_mc_events_total) > 0
+        for: 0m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
+        annotations:
+          new_errors: "{{ $value }}"
+      - alert: pcie aer errors
+        expr: increase(rasdaemon_aer_events_total) > 0
+        for: 0m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
+        annotations:
+          new_errors: "{{ $value }}"
   - name: smart
     rules:
       - alert: smart failure
-- 
2.39.5