Drop redundant alert that does nothing

[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 956c0d5b3a729935bedc1bbf087672c79ddf1180..598179d5cb4a6f5303d64f601766a61b5f65e61f 100644 (file)
--- a/cookbooks/prometheus/templates/default/alert_rules.yml.erb
+++ b/cookbooks/prometheus/templates/default/alert_rules.yml.erb
@@ -310,6 +310,20 @@ groups:
         for: 5m
         labels:
           alertgroup: "{{ $labels.site }}"
+      - alert: juniper laser receive power
+        expr: junos_interface_diagnostics_laser_rx_dbm < -12 and on (site, instance, name) junos_interface_admin_up == 1
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.site }}"
+        annotations:
+          power: "{{ $value }} dBm"
+      - alert: juniper laser transmit power
+        expr: junos_interface_diagnostics_laser_output_dbm < -8 and on (site, instance, name) junos_interface_admin_up == 1
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.site }}"
+        annotations:
+          power: "{{ $value }} dBm"
   - name: mail
     rules:
       - alert: exim down
@@ -402,6 +416,13 @@ groups:
           connections_used: "{{ $value | humanizePercentage }}"
   - name: network
     rules:
+      - alert: interface redundancy lost
+        expr: node_bonding_active < 2 and on (instance, master) label_replace(chef_network_interface{bond_mode="802.3ad"}, "master", "$1", "name", "(.*)")
+        for: 5m
+        labels:
+          alertgroup: "{{ $labels.instance }}"
+        annotations:
+          link_count: "{{ $value }}"
       - alert: interface transmit rate
         expr: rate(node_network_transmit_bytes_total[1m]) / node_network_speed_bytes > 0.99
         for: 5m
@@ -558,13 +579,6 @@ groups:
           alertgroup: "{{ $labels.instance }}"
         annotations:
           new_deadlocks: "{{ $value }}"
-      - alert: postgresql slow queries
-        expr: pg_slow_queries > 0
-        for: 5m
-        labels:
-          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="300"}) by (instance, server)
         for: 5m