X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/755f734fc7531c6f00fb5bd5067491c896bf97f4..eaf2ba261061613a6e23f6514afeb4f50817db66:/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 d717e4f0a..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
@@ -693,14 +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