X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/99423336378c46014559344bf30b8c6b5d8724ae..fa2ffc4b6ec0b9572dac9470bcfb474894b8a722:/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 9a9008ac2..92c38d5c5 100644
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -25,7 +25,7 @@ groups:
   - name: cpu
     rules:
       - alert: cpu pressure
-        expr: rate(node_pressure_memory_waiting_seconds_total[5m]) > 0.3
+        expr: rate(node_pressure_cpu_waiting_seconds_total[5m]) > 0.3
         for: 15m
         labels:
           alertgroup: "{{ $labels.instance }}"
@@ -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
@@ -95,7 +104,7 @@ groups:
           in_max_volts: "{{ with printf \"node_hwmon_in_max_volts{instance='%s',chip='%s',sensor='%s'}\" $labels.instance $labels.chip $labels.sensor | query }}{{ . | first | value | humanize }}V{{end}}"
   - name: io
     rules:
-      - alert: cpu pressure
+      - alert: io pressure
         expr: rate(node_pressure_io_waiting_seconds_total[5m]) > 0.6
         for: 60m
         labels:
@@ -130,6 +139,22 @@ groups:
         for: 5m
         labels:
           alertgroup: "{{ $labels.instance }}"
+  - name: mail
+    rules:
+      - 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