Merge remote-tracking branch 'tigerfell/pr257'

[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 30a14a471262afc886129a99075e15500452d144..28a5b311363a2fe2565bc92906a9dce803e6c39a 100644 (file)
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -115,7 +115,12 @@ groups:
         annotations:
           error_rate: "{{ $value | humanizePercentage }}"
       - alert: fastly healthcheck failing
-        expr: fastly_healthcheck_status == 0
+        expr: count(fastly_healthcheck_status == 0) > 0
+        for: 15m
+        labels:
+          alertgroup: fastly
+      - alert: fastly healthcheck failing
+        expr: count(fastly_healthcheck_status == 0) > 4
         for: 5m
         labels:
           alertgroup: fastly
@@ -447,7 +452,7 @@ groups:
         labels:
           alertgroup: "{{ $labels.instance }}"
       - alert: smart ssd wearout approaching
-        expr: smart_percentage_used >= 90
+        expr: smart_percentage_used >= 80
         for: 60m
         labels:
           alertgroup: "{{ $labels.instance }}"