X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/38df188a80a99f6d0bac9a88ce0454259dfa0f8b..d81b6d04058d580f4c0792726911d9837ead0f81:/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 7974547cb..956c0d5b3 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -52,6 +52,13 @@ groups:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           busy_workers: "{{ $value | humanizePercentage }}"
+      - alert: apache connection limit
+        expr: (apache_connections{state="total"} - on (instance) apache_connections{state="closing"}) / on (instance) (apache_server_limit * on (instance) (apache_threads_per_child + on (instance) (apache_async_request_worker_factor * on (instance) apache_workers{state="idle"} / on(instance) apache_processes{state="all"}))) > 0.8
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
+        annotations:
+          connections: "{{ $value | humanizePercentage }}"
   - name: chef
     rules:
       - alert: chef client not running
@@ -272,8 +279,22 @@ 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 > 0.5
+        expr: junos_route_engine_load_average_five / 2 > 0.5
         for: 5m
         labels:
           alertgroup: "{{ $labels.site }}"
@@ -679,14 +700,14 @@ 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