dns: update dnscontrol to 3.31.4

[chef.git] / 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 c34647330051da1d2f98ea06b90959724cc82ccd..b53007521e9c3df529ff600d122c5db4d04c217f 100644 (file)
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -4,19 +4,19 @@ groups:
   - name: amsterdam
     rules:
       - alert: pdu current draw
-        expr: rPDU2PhaseStatusCurrent{site="amsterdam",rPDU2PhaseStatusIndex="1"} / 10 > 10
+        expr: rPDU2PhaseStatusCurrent{site="amsterdam",rPDU2PhaseStatusIndex="1"} / 10 > 28
         for: 6m
         labels:
           alertgroup: "amsterdam"
         annotations:
           current: "{{ $value | humanize }}A"
-      - alert: site current draw
-        expr: sum(rPDU2PhaseStatusCurrent{site="amsterdam",rPDU2PhaseStatusIndex="1"} / 10) > 13
+      - alert: site power
+        expr: sum(avg_over_time(rPDU2PhaseStatusApparentPower{site="amsterdam",rPDU2PhaseStatusIndex="1"}[1h]) / 100) > 3
         for: 6m
         labels:
           alertgroup: "amsterdam"
         annotations:
-          current: "{{ $value | humanize }}A"
+          current: "{{ $value | humanize }}kVA"
       - alert: site temperature
         expr: min(rPDU2SensorTempHumidityStatusTempC{site="amsterdam"}) / 10 < 18 or min(rPDU2SensorTempHumidityStatusTempC{site="amsterdam"}) / 10 > 26
         for: 6m
@@ -117,19 +117,19 @@ groups:
   - name: dublin
     rules:
       - alert: pdu current draw
-        expr: rPDU2PhaseStatusCurrent{site="dublin",rPDU2PhaseStatusIndex="1"} / 10 > 13
+        expr: rPDU2PhaseStatusCurrent{site="dublin",rPDU2PhaseStatusIndex="1"} / 10 > 28
         for: 6m
         labels:
           alertgroup: "dublin"
         annotations:
           current: "{{ $value | humanize }}A"
-      - alert: site current draw
-        expr: sum(rPDU2PhaseStatusCurrent{site="dublin",rPDU2PhaseStatusIndex="1"} / 10) > 17
+      - alert: site power
+        expr: sum(avg_over_time(rPDU2PhaseStatusApparentPower{site="dublin",rPDU2PhaseStatusIndex="1"}[1h]) / 100) > 4
         for: 6m
         labels:
           alertgroup: "dublin"
         annotations:
-          current: "{{ $value | humanize }}A"
+          current: "{{ $value | humanize }}kVA"
       - alert: site temperature
         expr: min(rPDU2SensorTempHumidityStatusTempC{site="dublin"}) / 10 < 18 or min(rPDU2SensorTempHumidityStatusTempC{site="dublin"}) / 10 > 26
         for: 6m
@@ -525,6 +525,13 @@ groups:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           queries: "{{ $value }}"
+      - alert: postgresql idle transactions
+        expr: sum(pg_process_idle_seconds_count{state="idle in transaction"}) by (instance, server) > sum(pg_process_idle_seconds_bucket{state="idle in transaction",le="120"}) by (instance, server)
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
+        annotations:
+          queries: "{{ $value }}"
   - name: prometheus
     rules:
       - alert: prometheus configuration error
@@ -544,6 +551,11 @@ groups:
         for: 5m
         labels:
           alertgroup: "{{ $labels.instance }}"
+      - alert: raid controller battery recharging
+        expr: ohai_controller_info{battery_status="recharging"} > 0
+        for: 4h
+        labels:
+          alertgroup: "{{ $labels.instance }}"
       - alert: raid array degraded
         expr: ohai_array_info{status="degraded"} > 0
         for: 5m
@@ -645,6 +657,29 @@ groups:
         for: 0m
         labels:
           alertgroup: "{{ $labels.instance }}"
+  - name: taginfo
+    rules:
+      - alert: taginfo planet age
+        expr: time() - taginfo_data_from_seconds > 129600
+        for: 0m
+        labels:
+          alertgroup: taginfo
+        annotations:
+          age: "{{ $value | humanizeDuration }}"
+      - alert: taginfo database age
+        expr: time() - taginfo_database_update_finish_seconds > 129600
+        for: 0m
+        labels:
+          alertgroup: taginfo
+        annotations:
+          age: "{{ $value | humanizeDuration }}"
+      - alert: taginfo database size
+        expr: abs(delta(taginfo_database_size_bytes[30m])) / taginfo_database_size_bytes > 0.1
+        for: 30m
+        labels:
+          alertgroup: taginfo
+        annotations:
+          size_change: "{{ $value | humanizePercentage }}"
   - name: tile
     rules:
       - alert: renderd replication delay