apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
  name: ory-alerts
  namespace: ory
  labels:
    role: alert-rules
spec:
  groups:
    - name: ory
      rules:
        - alert: HydraDown
          expr: up{job=~".*hydra.*"} == 0
          for: 2m
          labels:
            severity: critical
          annotations:
            summary: "Hydra is down"
            description: "Hydra instance {{ $labels.namespace }}/{{ $labels.pod }} is down."

        - alert: KratosDown
          expr: up{job=~".*kratos.*"} == 0
          for: 2m
          labels:
            severity: critical
          annotations:
            summary: "Kratos is down"
            description: "Kratos instance {{ $labels.namespace }}/{{ $labels.pod }} is down."

        - alert: HydraHighErrorRate
          expr: sum(rate(http_requests_total{job=~".*hydra.*",code=~"5.."}[5m])) / sum(rate(http_requests_total{job=~".*hydra.*"}[5m])) > 0.05
          for: 5m
          labels:
            severity: warning
          annotations:
            summary: "Hydra has a high HTTP error rate"
            description: "Hydra 5xx error rate is {{ $value | humanizePercentage }}."

        - alert: KratosHighErrorRate
          expr: sum(rate(http_requests_total{job=~".*kratos.*",code=~"5.."}[5m])) / sum(rate(http_requests_total{job=~".*kratos.*"}[5m])) > 0.05
          for: 5m
          labels:
            severity: warning
          annotations:
            summary: "Kratos has a high HTTP error rate"
            description: "Kratos 5xx error rate is {{ $value | humanizePercentage }}."