water-management-system/deploy/production/monitoring/alertmanager.yml

# ============================================================
# AlertManager 配置 - 告警路由与通知
# ============================================================

global:
  resolve_timeout: 5m

# ==================== 告警模板 ====================
templates:
  - '/etc/alertmanager/templates/*.tmpl'

# ==================== 路由规则 ====================
route:
  group_by: ['alertname', 'job']
  group_wait: 30s
  group_interval: 5m
  repeat_interval: 4h
  receiver: 'wecom'

  routes:
    # 严重告警 - 立即通知
    - match:
        severity: critical
      receiver: 'wecom-critical'
      group_wait: 10s
      repeat_interval: 1h

    # 普通告警
    - match:
        severity: warning
      receiver: 'wecom'
      repeat_interval: 4h

# ==================== 接收器 ====================
receivers:
  # 企业微信通知 - 普通告警
  - name: 'wecom'
    webhook_configs:
      - url: 'http://wecom-alert-proxy:8080/alert'
        send_resolved: true
        http_config:
          follow_redirects: true

  # 企业微信通知 - 严重告警
  - name: 'wecom-critical'
    webhook_configs:
      - url: 'http://wecom-alert-proxy:8080/alert-critical'
        send_resolved: true
        http_config:
          follow_redirects: true

# ==================== 告警抑制 ====================
inhibit_rules:
  # 同一告警，critical 触发时抑制 warning
  - source_match:
      severity: 'critical'
    target_match:
      severity: 'warning'
    equal: ['alertname', 'instance']

  # 服务宕机时抑制该服务的其他告警
  - source_match:
      alertname: 'ServiceDown'
    target_match_re:
      alertname: 'ServiceSlowResponse|HighErrorRate'
    equal: ['job']