rules.example

groups:
- name: cnc_detect
  rules:
  - alert: node_offline
    expr: delta(cnc_health_detect[30m]) == 0
    for: 30m
    labels:
      severity: "warning"
    annotations:
      description: "Instance {{ $labels.instance }} is offline"

  - alert: pollCrash_15s
    expr: sum(count_over_time(cnc_http_detect{status!="200"}[1m]))/sum(count_over_time(cnc_http_detect{}[1m])) == 1
    for: 1m
    labels:
      severity: "warning"
    annotations:
      description: "Instance {{ $labels.instance }} pollCrash_15s"

  - alert: qos_down_vod_15s
    expr: (count_over_time(cnc_http_detect{job="cnc_demand_elapsed_detect_100k",health="qos_down_vod_15s"}[3m]) + count_over_time(cnc_http_detect{job="cnc_demand_elapsed_detect_100k",status!="200"}[3m]))/count_over_time(cnc_http_detect{job="cnc_demand_elapsed_detect_100k"}[3m]) > 0.8
    for: 3m
    labels:
      severity: "warning"
    annotations:
      description: "Instance {{ $labels.instance }} qos_down_vod_15s"

  - alert: pic_qos_down
    expr: (count_over_time(cnc_http_detect{job="cnc_http_1_elapsed_detect_8B",health="pic_qos_down"}[3m]) + count_over_time(cnc_http_detect{job="cnc_http_1_elapsed_detect_8B",status!="200"}[3m])) > 2 or (count_over_time(cnc_http_detect{job="cnc_http_2_elapsed_detect_8B",health="pic_qos_down"}[3m]) + count_over_time(cnc_http_detect{job="cnc_http_2_elapsed_detect_8B",status!="200"}[3m])) > 2
    for: 3m
    labels:
      severity: "warning"
    annotations:
      description: "Instance {{ $labels.instance }} pic_qos_down"

  - alert: livePollCrash_15s
    expr: (count_over_time(cnc_http_detect{job="cnc_live_elapsed_detect_100k",health="livePollCrash_15s"}[1m]) + count_over_time(cnc_http_detect{job="cnc_live_elapsed_detect_100k",status!="200"}[1m]))/sum(count_over_time(cnc_http_detect{}[1m])) == 1
    for: 1m
    labels:
      severity: "warning"
    annotations:
      description: "Instance {{ $labels.instance }} livePollCrash_15s"