# Starter thresholds: tune to the measured baseline for this cluster.
groups:
  - name: wukongim-starter
    rules:
      - alert: WuKongIMScrapeFailed
        expr: up{job="wukongim"} == 0
        for: 2m
        labels:
          severity: warning
        annotations:
          summary: "WuKongIM metrics collection failed"
          description: "Check connectivity and /readyz for {{ $labels.instance }}."
      - alert: WuKongIMSendFailures
        expr: sum by (cluster, node_id, reason) (increase(wukongim_gateway_sendacks_total{job="wukongim",reason!="success"}[5m])) > 0
        for: 2m
        labels:
          severity: warning
        annotations:
          summary: "WuKongIM returned unsuccessful SENDACKs"
          description: "Check reason {{ $labels.reason }} on node {{ $labels.node_id }}."
      - alert: WuKongIMDeliveryQueueHigh
        expr: |
          (
            wukongim_delivery_recipient_worker_queue_depth{job="wukongim"}
            / clamp_min(wukongim_delivery_recipient_worker_queue_capacity{job="wukongim"}, 1)
          ) > 0.8
          and
          wukongim_delivery_recipient_worker_queue_capacity{job="wukongim"} > 0
        for: 10m
        labels:
          severity: warning
        annotations:
          summary: "WuKongIM delivery queue has remained above 80%"
          description: "Check delivery workers and downstream nodes before increasing queue capacity."