]> git.openstreetmap.org Git - chef.git/blobdiff - cookbooks/prometheus/templates/default/alert_rules.yml.erb
Change default postgres version for tile servers to 14
[chef.git] / cookbooks / prometheus / templates / default / alert_rules.yml.erb
index 2678f5e1a26aa4ceb2832bdcde1b837b6ffb0fbf..77400969edb72db718778b5737f97c3228b78d53 100644 (file)
@@ -272,6 +272,20 @@ groups:
           alertgroup: "{{ $labels.instance }}"
   - name: juniper
     rules:
+      - alert: juniper red alarms
+        expr: juniper_alarms_red_count > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.site }}"
+        annotations:
+          alarm_count: "{{ $value }} alarms"
+      - alert: juniper yellow alarms
+        expr: juniper_alarms_yellow_count > 0
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.site }}"
+        annotations:
+          alarm_count: "{{ $value }} alarms"
       - alert: juniper cpu alarm
         expr: junos_route_engine_load_average_five / 2 > 0.5
         for: 5m
@@ -679,14 +693,14 @@ groups:
   - name: taginfo
     rules:
       - alert: taginfo planet age
-        expr: time() - taginfo_data_from_seconds > 129600
+        expr: time() - taginfo_data_from_seconds > 129600 and on (instance) chef_role{name="taginfo"}
         for: 0m
         labels:
           alertgroup: taginfo
         annotations:
           age: "{{ $value | humanizeDuration }}"
       - alert: taginfo database age
-        expr: time() - taginfo_database_update_finish_seconds > 129600
+        expr: time() - taginfo_database_update_finish_seconds > 129600 and on (instance) chef_role{name="taginfo"}
         for: 0m
         labels:
           alertgroup: taginfo