Merge interface families

[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 848123602032a5cca93197285fa244009bb9022c..cb8d91e99444379eb24b51336e86b4f24c9373ea 100644 (file)
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -105,6 +105,15 @@ groups:
           alertgroup: database
         annotations:
           delay: "{{ $value | humanizeDuration }}"
+  - name: discourse
+    rules:
+      - alert: discourse job failure rate
+        expr: rate(discourse_job_failures[5m]) > 0
+        for: 5m
+        labels:
+          alertgroup: discourse
+        annotations:
+          failure_rate: "{{ $value }} jobs/s"
   - name: dublin
     rules:
       - alert: pdu current draw
@@ -569,12 +578,21 @@ groups:
         labels:
           alertgroup: "{{ $labels.instance }}"
       - alert: smart ssd wearout approaching
-        expr: smart_percentage_used >= 80
+        expr: smart_percentage_used / 100 >= 0.8
         for: 60m
         labels:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           percentage_used: "{{ $value | humanizePercentage }}"
+  - name: smokeping
+    rules:
+      - alert: packet loss
+        expr: 1 - (rate(smokeping_response_duration_seconds_count[5m]) / rate(smokeping_requests_total[5m])) > 0.02
+        for: 10m
+        labels:
+          alertgroup: smokeping
+        annotations:
+          loss_rate: "{{ $value | humanizePercentage }}"
   - name: snmp
     rules:
       - alert: snmp pdus missing