# PrometheusRule for Frappe Operator
# Apply with: kubectl apply -f docs/alert-rules.yaml
# Requires Prometheus Operator (monitoring.coreos.com)
apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
  name: frappe-operator-alerts
  namespace: frappe-operator-system
  labels:
    prometheus: k8s
    role: alert-rules
spec:
  groups:
  - name: frappe-operator
    rules:
    - alert: FrappeOperatorDown
      expr: absent(up{job="frappe-operator"}) == 1
      for: 5m
      labels:
        severity: critical
      annotations:
        summary: "Frappe Operator is down"
        description: "Frappe Operator has been down for more than 5 minutes"

    - alert: FrappeOperatorHighErrorRate
      expr: rate(frappe_operator_reconciliation_errors_total[5m]) > 0.1
      for: 5m
      labels:
        severity: warning
      annotations:
        summary: "High reconciliation error rate"
        description: "Controller {{ $labels.controller }} has error rate of {{ $value }} errors/sec"

    - alert: FrappeOperatorSlowReconciliation
      expr: histogram_quantile(0.95, rate(frappe_operator_reconciliation_duration_seconds_bucket[5m])) > 60
      for: 10m
      labels:
        severity: warning
      annotations:
        summary: "Slow reconciliation detected"
        description: "Controller {{ $labels.controller }} p95 reconciliation time is {{ $value }}s"

    - alert: FrappeSiteNotReady
      expr: kube_customresource_frappesite_status_phase{phase!="Ready"} == 1
      for: 15m
      labels:
        severity: warning
      annotations:
        summary: "FrappeSite not ready"
        description: "Site {{ $labels.name }} in namespace {{ $labels.namespace }} is not ready"

    - alert: FrappeBenchNotReady
      expr: kube_customresource_frappebench_status_phase{phase!="Ready"} == 1
      for: 15m
      labels:
        severity: warning
      annotations:
        summary: "FrappeBench not ready"
        description: "Bench {{ $labels.name }} in namespace {{ $labels.namespace }} is not ready"

  - name: frappe-workloads
    rules:
    - alert: FrappeGunicornRestarts
      expr: increase(kube_pod_container_status_restarts_total{container="gunicorn"}[1h]) > 3
      labels:
        severity: warning
      annotations:
        summary: "Gunicorn container restarting"
        description: "Gunicorn container in pod {{ $labels.pod }} has restarted {{ $value }} times in the last hour"

    - alert: FrappeWorkerQueueHigh
      expr: frappe_worker_queue_length > 100
      for: 10m
      labels:
        severity: warning
      annotations:
        summary: "Worker queue is backing up"
        description: "Queue {{ $labels.queue }} has {{ $value }} pending jobs"

    - alert: FrappeDatabaseConnectionsHigh
      expr: mysql_global_status_threads_connected / mysql_global_variables_max_connections > 0.8
      for: 5m
      labels:
        severity: warning
      annotations:
        summary: "Database connections near limit"
        description: "{{ $value | humanizePercentage }} of max connections in use"