KEDA Prometheus Alert Rules

groups:
- name: EmbeddedExporter
  rules:
    - alert: KedaScaledobjectErrors
      expr: 'increase(keda_scaled_object_errors_total[5m]) > 0'
      for: 5m
      labels:
        severity: warning
      annotations:
        summary: KEDA ScaledObject errors (instance {{ $labels.instance }})
        description: "ScaledObject {{ $labels.namespace }}/{{ $labels.scaledObject }} keeps failing to read its scalers ({{ $value | printf \"%.0f\" }} errors in 5m). The workload is not scaled on fresh data; check the KEDA operator logs and the scaler source (auth, address, query).\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
    
    - alert: KedaScaledjobErrors
      expr: 'increase(keda_scaled_job_errors_total[5m]) > 0'
      for: 5m
      labels:
        severity: warning
      annotations:
        summary: KEDA ScaledJob errors (instance {{ $labels.instance }})
        description: "ScaledJob {{ $labels.namespace }}/{{ $labels.scaledJob }} keeps failing to read its scalers ({{ $value | printf \"%.0f\" }} errors in 5m). No new Jobs are created on fresh data; check the KEDA operator logs and the scaler source.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
    
      # Threshold of 5s is arbitrary. Adjust it to the polling interval of your ScaledObjects and to how slow the scaler source (Prometheus query, cloud API) normally is.
    - alert: KedaScalerMetricsLatencyHigh
      expr: 'keda_scaler_metrics_latency_seconds > 5'
      for: 5m
      labels:
        severity: warning
      annotations:
        summary: KEDA scaler metrics latency high (instance {{ $labels.instance }})
        description: "Scaler {{ $labels.scaler }} of {{ $labels.namespace }}/{{ $labels.scaledObject }} takes {{ $value | humanizeDuration }} to return its metric. Slow scaler sources delay scaling decisions.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
    
      # Threshold of 30s is arbitrary. Keep it above the polling interval of your ScaledObjects.
    - alert: KedaScalingLoopDelayed
      expr: 'keda_internal_scale_loop_latency_seconds > 30'
      for: 5m
      labels:
        severity: warning
      annotations:
        summary: KEDA scaling loop delayed (instance {{ $labels.instance }})
        description: "KEDA scaling loop for {{ $labels.type }} {{ $labels.namespace }}/{{ $labels.resource }} runs {{ $value | humanizeDuration }} later than scheduled. The operator is overloaded or blocked on slow scalers.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
    
    - alert: KedaCloudeventsEmissionFailing
      expr: 'increase(keda_cloudeventsource_events_emitted_total{state="failed"}[5m]) > 0'
      for: 0m
      labels:
        severity: warning
      annotations:
        summary: KEDA CloudEvents emission failing (instance {{ $labels.instance }})
        description: "KEDA cannot deliver CloudEvents from {{ $labels.namespace }}/{{ $labels.cloudeventsource }} to sink {{ $labels.eventsink }}.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"
    
    - alert: KedaScaledobjectPaused
      expr: 'keda_scaled_object_paused > 0'
      for: 24h
      labels:
        severity: info
      annotations:
        summary: KEDA ScaledObject paused (instance {{ $labels.instance }})
        description: "ScaledObject {{ $labels.namespace }}/{{ $labels.scaledObject }} has been paused for more than 24h, so its workload is not autoscaled.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"

Scrape the KEDA operator metrics endpoint (port 8080 by default).
KEDA sets its own namespace label. Without honor_labels, Prometheus renames it to exported_namespace.

wget https://raw.githubusercontent.com/samber/awesome-prometheus-alerts/refs/heads/master/dist/rules/keda/embedded-exporter.yml

- alert: KedaScaledobjectErrors
  expr: 'increase(keda_scaled_object_errors_total[5m]) > 0'
  for: 5m
  labels:
    severity: warning
  annotations:
    summary: KEDA ScaledObject errors (instance {{ $labels.instance }})
    description: "ScaledObject {{ $labels.namespace }}/{{ $labels.scaledObject }} keeps failing to read its scalers ({{ $value | printf \"%.0f\" }} errors in 5m). The workload is not scaled on fresh data; check the KEDA operator logs and the scaler source (auth, address, query).\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"

- alert: KedaScaledjobErrors
  expr: 'increase(keda_scaled_job_errors_total[5m]) > 0'
  for: 5m
  labels:
    severity: warning
  annotations:
    summary: KEDA ScaledJob errors (instance {{ $labels.instance }})
    description: "ScaledJob {{ $labels.namespace }}/{{ $labels.scaledJob }} keeps failing to read its scalers ({{ $value | printf \"%.0f\" }} errors in 5m). No new Jobs are created on fresh data; check the KEDA operator logs and the scaler source.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"

  # Threshold of 5s is arbitrary. Adjust it to the polling interval of your ScaledObjects and to how slow the scaler source (Prometheus query, cloud API) normally is.
- alert: KedaScalerMetricsLatencyHigh
  expr: 'keda_scaler_metrics_latency_seconds > 5'
  for: 5m
  labels:
    severity: warning
  annotations:
    summary: KEDA scaler metrics latency high (instance {{ $labels.instance }})
    description: "Scaler {{ $labels.scaler }} of {{ $labels.namespace }}/{{ $labels.scaledObject }} takes {{ $value | humanizeDuration }} to return its metric. Slow scaler sources delay scaling decisions.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"

  # Threshold of 30s is arbitrary. Keep it above the polling interval of your ScaledObjects.
- alert: KedaScalingLoopDelayed
  expr: 'keda_internal_scale_loop_latency_seconds > 30'
  for: 5m
  labels:
    severity: warning
  annotations:
    summary: KEDA scaling loop delayed (instance {{ $labels.instance }})
    description: "KEDA scaling loop for {{ $labels.type }} {{ $labels.namespace }}/{{ $labels.resource }} runs {{ $value | humanizeDuration }} later than scheduled. The operator is overloaded or blocked on slow scalers.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"

- alert: KedaCloudeventsEmissionFailing
  expr: 'increase(keda_cloudeventsource_events_emitted_total{state="failed"}[5m]) > 0'
  for: 0m
  labels:
    severity: warning
  annotations:
    summary: KEDA CloudEvents emission failing (instance {{ $labels.instance }})
    description: "KEDA cannot deliver CloudEvents from {{ $labels.namespace }}/{{ $labels.cloudeventsource }} to sink {{ $labels.eventsink }}.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"

- alert: KedaScaledobjectPaused
  expr: 'keda_scaled_object_paused > 0'
  for: 24h
  labels:
    severity: info
  annotations:
    summary: KEDA ScaledObject paused (instance {{ $labels.instance }})
    description: "ScaledObject {{ $labels.namespace }}/{{ $labels.scaledObject }} has been paused for more than 24h, so its workload is not autoscaled.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"