X-Git-Url: https://git.openstreetmap.org./chef.git/blobdiff_plain/9efef35f6b4ac2ccdd62a6deb57019a6ba14ce7e..030b284b1d91df3985bff215589989df9e6db0b7:/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 956c0d5b3..d8671699c 100644 --- 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