X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/6cdd946b4e56b4a6593a5f7df3051c7ba95bebeb..18131b23c3379c9bd43356bfa559f3ea7fc92685:/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 9a3609e03..be6a89b43 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -40,6 +40,15 @@ groups:
           alertgroup: database
         annotations:
           delay: "{{ $value | humanizeDuration }}"
+  - name: fastly
+    rules:
+      - alert: error rate
+        expr: sum(rate(fastly_rt_status_group_total{status_group="5xx"}[5m])) by (service_name, datacenter) / sum(rate(fastly_rt_status_group_total[5m])) by (service_name, datacenter) > 0.005
+        for: 15m
+        labels:
+          alertgroup: fastly
+        annotations:
+          error_rate: "{{ $value | humanizePercentage }}"
   - name: filesystem
     rules:
       - alert: readonly filesystem
@@ -132,13 +141,20 @@ groups:
           alertgroup: "{{ $labels.instance }}"
   - name: mail
     rules:
-      - alert: mail queue length
+      - alert: exim queue length
         expr: exim_queue > exim_queue_limit
         for: 60m
         labels:
           alertgroup: mail
         annotations:
           queue_length: "{{ $value }}"
+      - alert: mailman queue length
+        expr: mailman_queue_length > 200
+        for: 60m
+        labels:
+          alertgroup: mail
+        annotations:
+          queue_length: "{{ $value }}"
   - name: mdadm
     rules:
       - alert: mdadm array inactive
@@ -394,3 +410,10 @@ groups:
           alertgroup: web
         annotations:
           error_rate: "{{ $value | humanizePercentage }}"
+      - alert: job processing rate
+        expr: rate(pg_stat_user_tables_n_tup_del{datname="openstreetmap",relname="delayed_jobs"}[5m]) / rate(pg_stat_user_tables_n_tup_ins{datname="openstreetmap",relname="delayed_jobs"}[5m]) < 0.9 and ignoring(job, name, datname, relname, schemaname, server) chef_role{name="db-master"} == 1
+        for: 5m
+        labels:
+          alertgroup: web
+        annotations:
+          job_processing_rate: "{{ $value | humanizePercentage }}"