podman: Increase podman-auto-update.service frequency

[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 34a33d374ea1de8d9a385c724f05a573a8ea9a08..367e07650255de3383d13955ea1b73325bf3fc8d 100644 (file)
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -337,6 +337,20 @@ groups:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           new_oom_kills: "{{ $value }}"
+  - name: mysql
+    rules:
+      - alert: mysql down
+        expr: mysql_up == 0
+        for: 1m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
+      - alert: mysql connection limit
+        expr: mysql_global_status_max_used_connections / mysql_global_variables_max_connections > 0.8
+        for: 1m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
+        annotations:
+          connections_used: "{{ $value | humanizePercentage }}"
   - name: network
     rules:
       - alert: interface transmit rate
@@ -541,7 +555,7 @@ groups:
         labels:
           alertgroup: "{{ $labels.instance }}"
         annotations:
-          new_ercrors: "{{ $value }}"
+          new_errors: "{{ $value }}"
   - name: smart
     rules:
       - alert: smart failure
@@ -556,6 +570,15 @@ groups:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           percentage_used: "{{ $value | humanizePercentage }}"
+  - name: snmp
+    rules:
+      - alert: snmp pdus missing
+        expr: max_over_time(snmp_scrape_pdus_returned[1d]) - snmp_scrape_pdus_returned > 0
+        for: 15m
+        labels:
+          alertgroup: snmp
+        annotations:
+          missing_pdus: "{{ $value }}"
   - name: ssl
     rules:
       - alert: ssl certificate probe failed
@@ -615,6 +638,13 @@ groups:
           alertgroup: tile
         annotations:
           miss_rate: "{{ $value | humanizePercentage }}"
+      - alert: tile render rate
+        expr: sum(rate(renderd_zoom_metatiles_total[5m])) by (instance) == 0
+        for: 15m
+        labels:
+          alertgroup: tile
+        annotations:
+          render_rate: "{{ $value }} tiles/s"
   - name: time
     rules:
       - alert: clock not synchronising