diff --git a/kubernetes/helm_charts/monitoring/alertrules/templates/promrulesNode.yml b/kubernetes/helm_charts/monitoring/alertrules/templates/promrulesNode.yml index d5fdeb67d5..f743a8ea8e 100644 --- a/kubernetes/helm_charts/monitoring/alertrules/templates/promrulesNode.yml +++ b/kubernetes/helm_charts/monitoring/alertrules/templates/promrulesNode.yml @@ -13,7 +13,7 @@ spec: - name: alertrules.nodes rules: - alert: high_cpu_usage_on_node_warning - expr: (avg by (instance) (irate(node_cpu_seconds_total{job="vm-node-exporter",mode="idle"}[5m])) * 100) >= {{ .Values.node_cpu_usage_percentage_threshold_Warning }} and (avg by (instance) (irate(node_cpu_seconds_total{job="vm-node-exporter",mode="idle"}[5m])) * 100) < {{ .Values.node_cpu_usage_percentage_threshold_Critical }} + expr: (avg by (instance) (irate(node_cpu_seconds_total{job="vm-node-exporter",mode!="idle"}[5m])) * 100) >= {{ .Values.node_cpu_usage_percentage_threshold_Warning }} and (avg by (instance) (irate(node_cpu_seconds_total{job="vm-node-exporter",mode!="idle"}[5m])) * 100) < {{ .Values.node_cpu_usage_percentage_threshold_Critical }} for: 1m labels: severity: warning @@ -21,7 +21,7 @@ spec: message: {{`'{{ $labels.instance }} is using a LOT of CPU. CPU usage is {{ humanize $value}}%.'`}} summary: {{`'HIGH CPU USAGE warning ON {{ $labels.instance }}'`}} - alert: high_cpu_usage_on_node_critical - expr: (avg by (instance) (irate(node_cpu_seconds_total{job="vm-node-exporter",mode="idle"}[5m])) * 100) >= {{ .Values.node_cpu_usage_percentage_threshold_Critical }} and (avg by (instance) (irate(node_cpu_seconds_total{job="vm-node-exporter",mode="idle"}[5m])) * 100) < {{ .Values.node_cpu_usage_percentage_threshold_Fatal }} + expr: (avg by (instance) (irate(node_cpu_seconds_total{job="vm-node-exporter",mode!="idle"}[5m])) * 100) >= {{ .Values.node_cpu_usage_percentage_threshold_Critical }} and (avg by (instance) (irate(node_cpu_seconds_total{job="vm-node-exporter",mode!="idle"}[5m])) * 100) < {{ .Values.node_cpu_usage_percentage_threshold_Fatal }} for: 1m labels: severity: critical @@ -29,7 +29,7 @@ spec: message: {{`'{{ $labels.instance }} is using a LOT of CPU. CPU usage is {{ humanize $value}}%.'`}} summary: {{`'HIGH CPU USAGE warning ON {{ $labels.instance }}'`}} - alert: high_cpu_usage_on_node_fatal - expr: (avg by (instance) (irate(node_cpu_seconds_total{job="vm-node-exporter",mode="idle"}[5m])) * 100) >= {{ .Values.node_cpu_usage_percentage_threshold_Fatal }} + expr: (avg by (instance) (irate(node_cpu_seconds_total{job="vm-node-exporter",mode!="idle"}[5m])) * 100) >= {{ .Values.node_cpu_usage_percentage_threshold_Fatal }} for: 1m labels: severity: fatal