---
apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
  labels:
    app: traefik
  name: traefik.rules
  namespace: networking
spec:
  groups:
    - name: traefik.rules
      rules:
        - alert: TraefikAbsent
          annotations:
            summary: "Traefik has disappeared from Prometheus service discovery."
            description: "Ingresses will be down until the Traefik reverse proxy is back up."
          expr: |
            absent(up{job="traefik"})
          for: 5m
          labels:
            severity: critical
        - alert: TraefikConfigError
          annotations:
            summary: "Traefik config error."
            description:
              "Traefik has failed to load the config file. Check Traefik
              logs for exact parsing error."
          expr: |
            traefik_config_last_reload_failure{job="traefik"} == 1
          for: 5m
          labels:
            severity: critical
        - alert: TraefikHighHttp4xxErrorRateService
          annotations:
            summary: "Traefik has a high HTTP 4xx error rate."
            description:
              "Traefik is reporting {{ $value | humanizePercentage }} of 4xx
              errors on {{ $labels.exported_service }}"
          expr: |
            sum(rate(traefik_service_requests_total{code=~"4.*"}[1m])) by (exported_service)
              /
            sum(rate(traefik_service_requests_total[1m])) by (exported_service)
              > .10
          for: 5m
          labels:
            severity: critical
        - alert: TraefikHighHttp5xxErrorRateService
          annotations:
            summary: "Traefik has a high HTTP 5xx error rate."
            description:
              "Traefik is reporting {{ $value | humanizePercentage }} of 5xx
              errors on {{ $labels.exported_service }}"
          expr: |
            sum(rate(traefik_service_requests_total{code=~"5.*"}[1m])) by (exported_service)
              /
            sum(rate(traefik_service_requests_total[1m])) by (exported_service)
              > .10
          for: 5m
          labels:
            severity: critical
        - alert: TraefikTooManyRequest
          annotations:
            summary: "Traefik has too many open connections"
            description:
              "Traefik is reporting {{ $value }} of open connections on entrypoint
              {{ $labels.entrypoint }}"
          expr: |
            avg(traefik_entrypoint_open_connections{job="traefik"})
              > 5
          for: 5m
          labels:
            severity: critical