Add alert rules for cisco switches

[chef.git] / cookbooks / prometheus / templates / default / alert_rules.yml.erb

diff --git a/cookbooks/prometheus/templates/default/alert_rules.yml.erb b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
index 4b5a0c60f2e14517c5d8b681001d01e382aa2d57..289b33f572059289ef5f8dd7dc7d8a27b3faee3d 100644 (file)
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -53,7 +53,7 @@ groups:
         annotations:
           busy_workers: "{{ $value | humanizePercentage }}"
       - alert: apache low request rate
-        expr: rate(apache_accesses_total[$__rate_interval]) / rate(apache_accesses_total[$__rate_interval] offset 1w) < 0.25 and rate(apache_accesses_total[1h] offset 1w) > 2
+        expr: rate(apache_accesses_total[5m]) / rate(apache_accesses_total[1h] offset 1w) < 0.25 and rate(apache_accesses_total[1h] offset 1w) > 2
         for: 15m
         labels:
           alertgroup: "{{ $labels.instance }}"
@@ -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