---
apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
  namespace: {{ .Release.Namespace }}
  name: {{ include "dell-exporter.fullname" . }}
  labels:
{{ include "dell-exporter.labels" . | indent 4 }}
spec:
  groups:
  - name: dell
    rules:
    - alert: DellSystemStatus
      expr: |
        dell_hw_system_status == 2
      for: 1m
      labels:
        severity: P4
      annotations:
        summary: "[`{{`{{$labels.instance}}`}}`] System warning"
        description: >
          The host `{{`{{$labels.instance}}`}}` is reporting a non-critical status,
          please investigate.  If this is the only firing alert, please add
          another one to handle the specific failure.

    - alert: DellSystemStatus
      expr: |
        dell_hw_system_status == 1
      for: 1m
      labels:
        severity: P3
      annotations:
        summary: "[`{{`{{$labels.instance}}`}}`] System failure"
        description: >
          The host `{{`{{$labels.instance}}`}}` is reporting a critical status,
          please investigate.  If this is the only firing alert, please add
          another one to handle the specific failure.

    - alert: DellFanFailure
      expr: |
        dell_hw_chassis_fan_status != 0
      for: 1m
      labels:
        severity: P4
      annotations:
        summary: "[`{{`{{$labels.instance}}`}}`] Fan failure"
        description: >
          The `{{`{{$labels.fan}}`}}` on host `{{`{{$labels.instance}}`}}` is reporting
          a failure, please replace it.

    - alert: DellMemoryFailure
      expr: |
        dell_hw_chassis_memory_status != 0
      for: 1m
      labels:
        severity: P4
      annotations:
        summary: "[`{{`{{$labels.instance}}`}}`] Memory failure"
        description: >
          The `{{`{{$labels.memory}}`}}` on host `{{`{{$labels.instance}}`}}` is reporting
          a failure, please replace it.

    - alert: DellPowerSupplyFailure
      expr: |
        dell_hw_ps_status != 0
      for: 1m
      labels:
        severity: P3
      annotations:
        summary: "[`{{`{{$labels.instance}}`}}`] Power failure"
        description: >
          The PSU `{{`{{$labels.id}}`}}` on host `{{`{{$labels.instance}}`}}` is reporting
          a failure, please investigate if power is lost or power supply
          requires replacement.

    - alert: DellHardwareStorageFailure
      expr: |
        dell_hw_storage_controller_status != 0
      for: 1m
      labels:
        severity: P3
      annotations:
        summary: "[`{{`{{$labels.instance}}`}}`] Storage failure"
        description: >
          The storage controller `{{`{{$labels.id}}`}}` on host `{{`{{$labels.instance}}`}}`
          is reporting a failure, please investigate inside the host and add the
          appropriate alerting rules if no alerts except this one have fired.

    - alert: DellChassisTemperature
      expr: |
        dell_hw_chassis_temps_reading > dell_hw_chassis_temps_max_warning
      labels:
        severity: P3
      annotations:
        summary: "[{{`{{$labels.instance}}`}}] {{`{{$labels.component}}`}}"
        description: >
          The component {{`{{$labels.component}}`}} is reporting temperatures
          of {{`{{$value}}`}} which is above the critical reading for
          {{`{{$labels.instance}}`}}.

    - alert: DellChassisTemperature
      expr: |
        dell_hw_chassis_temps_reading > dell_hw_chassis_temps_max_warning
      labels:
        severity: P4
      annotations:
        summary: "[{{`{{$labels.instance}}`}}] {{`{{$labels.component}}`}}"
        description: >
          The component {{`{{$labels.component}}`}} is reporting temperatures
          of {{`{{$value}}`}} which is above the warning reading for
          {{`{{$labels.instance}}`}}.