diff --git a/tmpl/alert-rules.yml b/tmpl/alert-rules.yml index 8c6a684..72d5c01 100644 --- a/tmpl/alert-rules.yml +++ b/tmpl/alert-rules.yml @@ -39,3 +39,12 @@ groups: a value of {{ $value }}. summary: Instance {{ $labels.instance }} filesystem usage is dangerously high + # Heavy CPU temperature + - alert: cpu_temp_threshold_exceeded + expr: avg(node_hwmon_temp_celsius{job="node"}) BY (instance) + > 50 + annotations: + description: This device's cpu temperature has exceeded the threshold with a value + of {{ $value }}. + summary: Instance {{ $labels.instance }} CPU temperature is dangerously high +