X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/e442c6a6365b6be91e350d9fe79d3a6227e7295c..49cdf148a0562915edfd17e9b4ab19c9cd7aa721:/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 23bb157e4..28a5b3113 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -1,13 +1,6 @@
 # DO NOT EDIT - This file is being maintained by Chef
 
 groups:
-  - name: alertmanager
-    rules:
-      - alert: prometheus target missing
-        expr: up == 0
-        for: 10m
-        labels:
-          alertgroup: "prometheus"
   - name: amsterdam
     rules:
       - alert: pdu current draw
@@ -122,7 +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
@@ -422,6 +420,18 @@ groups:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           queries: "{{ $value }}"
+  - name: prometheus
+    rules:
+      - alert: prometheus configuration error
+        expr: prometheus_config_last_reload_successful == 0
+        for: 10m
+        labels:
+          alertgroup: "prometheus"
+      - alert: prometheus target missing
+        expr: up == 0
+        for: 10m
+        labels:
+          alertgroup: "prometheus"
   - name: raid
     rules:
       - alert: raid array degraded
@@ -442,7 +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 }}"