Reduce sensitivity of CPU pressure alerts

[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 f794381820a4c7971e566990c264b877cf832aa4..c34647330051da1d2f98ea06b90959724cc82ccd 100644 (file)
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -90,8 +90,8 @@ groups:
   - name: cpu
     rules:
       - alert: cpu pressure
   - name: cpu
     rules:
       - alert: cpu pressure

-        expr: rate(node_pressure_cpu_waiting_seconds_total[5m]) > 0.6
-        for: 15m
+        expr: rate(node_pressure_cpu_waiting_seconds_total[5m]) > 0.75
+        for: 60m

         labels:
           alertgroup: "{{ $labels.instance }}"
         annotations:
         labels:
           alertgroup: "{{ $labels.instance }}"
         annotations:


@@ -587,8 +587,8 @@ groups:
   - name: smokeping
     rules:
       - alert: packet loss
   - name: smokeping
     rules:
       - alert: packet loss

-        expr: 1 - (rate(smokeping_response_duration_seconds_count[5m]) / rate(smokeping_requests_total[5m])) > 0
-        for: 5m
+        expr: 1 - (rate(smokeping_response_duration_seconds_count[5m]) / rate(smokeping_requests_total[5m])) > 0.02
+        for: 10m

         labels:
           alertgroup: smokeping
         annotations:
         labels:
           alertgroup: smokeping
         annotations: