Add new file

prometheus-node-exporter-rules.yaml

+apiVersion: monitoring.coreos.com/v1
+kind: PrometheusRule
+metadata:
+  labels:
+    prometheus: k8s
+    role: alert-rules
+  name: prometheus-node-exporter-rules
+  namespace: monitoring
+spec:
+  groups:
+  - name: node-exporter
+    rules:
+    - alert: NodeCPUUtilizaionHigh
+      annotations:
+        message: CPU Utilizaion on node {{$labels.instance}} is {{ $value | printf "%.2f" }}%.
+        runbook_url: RUNBOOK_URL
+      expr: |
+        avg by(instance) (sum by(instance, cpu) (irate(node_cpu_seconds_total{job="node-exporter", mode!="idle"}[2m]))) * 100 > 40
+      for: 2m
+      labels:
+        severity: warning
+    - alert: NodeCPUUtilizaionHigh
+      annotations:
+        message: CPU Utilizaion on node {{$labels.instance}} is {{ $value | printf "%.2f" }}%.
+        runbook_url: RUNBOOK_URL
+      expr: |
+        avg by(instance) (sum by(instance, cpu) (irate(node_cpu_seconds_total{job="node-exporter", mode!="idle"}[2m]))) * 100 > 60
+      for: 2m
+      labels:
+        severity: critical
+    - alert: NodeCPULoad5High
+      annotations:
+        message: CPU Load5 per core on node {{$labels.instance}} is {{ $value | printf "%.2f" }}.
+        runbook_url: RUNBOOK_URL
+      expr: |
+        (node_load5{job="node-exporter"})
+        / on(instance) (count by(instance) (sum by(instance, cpu) (node_cpu_seconds_total{job="node-exporter"}))) > 0.6
+      for: 2m
+      labels:
+        severity: warning
+    - alert: NodeCPULoad5High
+      annotations:
+        message: CPU Load5 per core on node {{$labels.instance}} is {{ $value | printf "%.2f" }}.
+        runbook_url: RUNBOOK_URL
+      expr: |
+        (node_load5{job="node-exporter"})
+        / on(instance) (count by(instance) (sum by(instance, cpu) (node_cpu_seconds_total{job="node-exporter"}))) > 0.8
+      for: 2m
+      labels:
+        severity: critical
+    - alert: NodeMemUsageHigh
+      annotations:
+        message: Memory Usage on node {{$labels.instance}} is {{ $value | printf "%.2f" }}%.
+        runbook_url: RUNBOOK_URL
+      expr: |
+        (node_memory_MemTotal_bytes{job="node-exporter"}
+        - on(instance) node_memory_MemFree_bytes{job="node-exporter"}
+        - on(instance) node_memory_Buffers_bytes{job="node-exporter"}
+        - on(instance) node_memory_Cached_bytes{job="node-exporter"})
+        / on(instance) node_memory_MemTotal_bytes{job="node-exporter"}
+        * 100 > 80
+      for: 2m
+      labels:
+        severity: warning
+    - alert: NodeMemUsageHigh
+      annotations:
+        message: Memory Usage on node {{$labels.instance}} is {{ $value | printf "%.2f" }}%.
+        runbook_url: RUNBOOK_URL
+      expr: |
+        (node_memory_MemTotal_bytes{job="node-exporter"}
+        - on(instance) node_memory_MemFree_bytes{job="node-exporter"}
+        - on(instance) node_memory_Buffers_bytes{job="node-exporter"}
+        - on(instance) node_memory_Cached_bytes{job="node-exporter"})
+        / on(instance) node_memory_MemTotal_bytes{job="node-exporter"}
+        * 100 > 90
+      for: 2m
+      labels:
+        severity: critical
+    - alert: NodeRxUsageHigh
+      annotations:
+        message: Network Rx on node {{$labels.instance}} is {{ $value | printf "%.2f" }}KB/S.
+        runbook_url: RUNBOOK_URL
+      expr: |
+        irate(node_network_receive_bytes_total{job="node-exporter", device="eth0"}[2m]) / 1024 > 4096
+      for: 2m
+      labels:
+        severity: warning
+    - alert: NodeRxUsageHigh
+      annotations:
+        message: Network Rx on node {{$labels.instance}} is {{ $value | printf "%.2f" }}KB/S.
+        runbook_url: RUNBOOK_URL
+      expr: |
+        irate(node_network_receive_bytes_total{job="node-exporter", device="eth0"}[2m]) / 1024 > 8192
+      for: 2m
+      labels:
+        severity: critical
+    - alert: NodeTxUsageHigh
+      annotations:
+        message: Network Tx on node {{$labels.instance}} is {{ $value | printf "%.2f" }}KB/S.
+        runbook_url: RUNBOOK_URL
+      expr: |
+        irate(node_network_transmit_bytes_total{job="node-exporter", device="eth0"}[2m]) / 1024 > 4096
+      for: 2m
+      labels:
+        severity: warning
+    - alert: NodeTxUsageHigh
+      annotations:
+        message: Network Tx on node {{$labels.instance}} is {{ $value | printf "%.2f" }}KB/S.
+        runbook_url: RUNBOOK_URL
+      expr: |
+        irate(node_network_transmit_bytes_total{job="node-exporter", device="eth0"}[2m]) / 1024 > 8192
+      for: 2m
+      labels:
+        severity: critical