From: Tom Hughes <tom@compton.nu>
Date: Thu, 21 Jul 2022 16:30:14 +0000 (+0100)
Subject: Add alert rules for cisco switches
X-Git-Url: https://git.openstreetmap.org./chef.git/commitdiff_plain/ab63149db6f2c9739f9dc39e0f0b18891f865690?ds=sidebyside

Add alert rules for cisco switches
---

diff --git a/cookbooks/prometheus/templates/default/alert_rules.yml.erb b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
index 0d045af71..289b33f57 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -68,6 +68,28 @@ groups:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           down_time: "{{ $value | humanizeDuration }}"
+  - name: cisco
+    rules:
+      - alert: cisco fan alarm
+        expr: rlPhdUnitEnvParamFan1Status{rlPhdUnitEnvParamFan1Status!="normal"} > 0 or rlPhdUnitEnvParamFan2Status{rlPhdUnitEnvParamFan2Status!="normal"} > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.site }}"
+      - alert: cisco temperature alarm
+        expr: rlPhdUnitEnvParamTempSensorStatus{rlPhdUnitEnvParamTempSensorStatus!="ok"} > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.site }}"
+      - alert: cisco main power alarm
+        expr: rlPhdUnitEnvParamMainPSStatus{rlPhdUnitEnvParamMainPSStatus!="normal"} > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.site }}"
+      - alert: cisco redundant power alarm
+        expr: rlPhdUnitEnvParamRedundantPSStatus{rlPhdUnitEnvParamRedundantPSStatus!="normal"} > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.site }}"
   - name: cpu
     rules:
       - alert: cpu pressure