community: bump discourse version

[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 20136a7715f6ae26b4ad14e0a52a1cf0792d915f..a0cea579289fde937a6e5947704aa618a1b5b65d 100644 (file)
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -25,14 +25,14 @@ groups:
         annotations:
           current: "{{ $value | humanize }}kVA"
       - alert: site temperature
         annotations:
           current: "{{ $value | humanize }}kVA"
       - alert: site temperature

-        expr: min(rPDU2SensorTempHumidityStatusTempC{site="amsterdam"}) / 10 < 18 or min(rPDU2SensorTempHumidityStatusTempC{site="amsterdam"}) / 10 > 26
+        expr: min(rPDU2SensorTempHumidityStatusTempC{site="amsterdam"}) / 10 < 15 or min(rPDU2SensorTempHumidityStatusTempC{site="amsterdam"}) / 10 > 32

         for: 6m
         labels:
           alertgroup: "amsterdam"
         annotations:
           temperature: "{{ $value | humanize }}C"
       - alert: site humidity
         for: 6m
         labels:
           alertgroup: "amsterdam"
         annotations:
           temperature: "{{ $value | humanize }}C"
       - alert: site humidity

-        expr: max(rPDU2SensorTempHumidityStatusRelativeHumidity{site="amsterdam"}) / 100 < 0.25 or max(rPDU2SensorTempHumidityStatusRelativeHumidity{site="amsterdam"}) / 100 > 0.65
+        expr: max(rPDU2SensorTempHumidityStatusRelativeHumidity{site="amsterdam"}) / 100 < 0.08 or max(rPDU2SensorTempHumidityStatusRelativeHumidity{site="amsterdam"}) / 100 > 0.8

         for: 6m
         labels:
           alertgroup: "amsterdam"
         for: 6m
         labels:
           alertgroup: "amsterdam"


@@ -106,14 +106,14 @@ groups:
   - name: database
     rules:
       - alert: active rails queries
   - name: database
     rules:
       - alert: active rails queries

-        expr: sum(pg_stat_activity_count{instance="snap-01",datname="openstreetmap",usename="rails",state="active"}) > 50 and on (instance) chef_role{name="db-master"}
+        expr: sum(pg_stat_activity_count{datname="openstreetmap",usename="rails",state="active"}) by (instance) > 50 and on (instance) chef_role{name="db-master"}

         for: 5m
         labels:
           alertgroup: database
         annotations:
           queries: "{{ $value }}"
       - alert: active cgimap queries
         for: 5m
         labels:
           alertgroup: database
         annotations:
           queries: "{{ $value }}"
       - alert: active cgimap queries

-        expr: sum(pg_stat_activity_count{instance="snap-01",datname="openstreetmap",usename="cgimap",state="active"}) > 30 and on (instance) chef_role{name="db-master"}
+        expr: sum(pg_stat_activity_count{datname="openstreetmap",usename="cgimap",state="active"}) by (instance) > 30 and on (instance) chef_role{name="db-master"}

         for: 5m
         labels:
           alertgroup: database
         for: 5m
         labels:
           alertgroup: database


@@ -421,6 +421,13 @@ groups:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           connections_used: "{{ $value | humanizePercentage }}"
           alertgroup: "{{ $labels.instance }}"
         annotations:
           connections_used: "{{ $value | humanizePercentage }}"

+      - alert: mysql connection errors
+        expr: increase(mysql_global_status_connection_errors_total[1m]) > 0
+        for: 0m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
+        annotations:
+          error_count: "{{ $value }}"

   - name: network
     rules:
       - alert: interface redundancy lost
   - name: network
     rules:
       - alert: interface redundancy lost


@@ -605,6 +612,11 @@ groups:
         for: 10m
         labels:
           alertgroup: "prometheus"
         for: 10m
         labels:
           alertgroup: "prometheus"

+      - alert: node exporter text file scrape error
+        expr: node_textfile_scrape_error > 0
+        for: 10m
+        labels:
+          alertgroup: "prometheus"

   - name: raid
     rules:
       - alert: raid controller battery failed
   - name: raid
     rules:
       - alert: raid controller battery failed