nightingale/integrations/linux/alerts/linux_by_telegraf.json

[
    {
      "name": "Address is unreachable by PING, please pay attention - telegraf",
      "note": "",
      "severity": 1,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "ping_result_code != 0",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    },
    {
      "cate": "host",
      "datasource_ids": null,
      "name": "Lost connection with monitoring target - telegraf",
      "note": "",
      "prod": "host",
      "algorithm": "",
      "algo_params": null,
      "delay": 0,
      "severity": 0,
      "disabled": 0,
      "prom_for_duration": 0,
      "prom_ql": "",
      "rule_config": {
        "inhibit": false,
        "queries": [
          {
            "key": "all_hosts",
            "op": "==",
            "values": []
          }
        ],
        "triggers": [
          {
            "duration": 60,
            "severity": 2,
            "type": "target_miss"
          }
        ]
      },
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_stimes": [
        "00:00"
      ],
      "enable_etime": "23:59",
      "enable_etimes": [
        "23:59"
      ],
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "enable_days_of_weeks": [
        [
          "1",
          "2",
          "3",
          "4",
          "5",
          "6",
          "0"
        ]
      ],
      "enable_in_bg": 0,
      "notify_recovered": 1,
      "notify_channels": [],
      "notify_repeat_step": 60,
      "notify_max_number": 0,
      "recover_duration": 0,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": [],
      "annotations": {}
    },
    {
      "name": "Port detection failed, please pay attention - telegraf",
      "note": "",
      "severity": 1,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "net_response_result_code != 0",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    },
    {
      "name": "Machine load - high CPU, please pay attention - telegraf",
      "note": "",
      "severity": 3,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "cpu_usage_idle{cpu=\"cpu-total\"} < 25",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    },
    {
      "name": "Machine load - high memory, please pay attention - telegraf",
      "note": "",
      "severity": 2,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "mem_available_percent < 25",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    },
    {
      "name": "Hard disk - IO is very busy - telegraf",
      "note": "",
      "severity": 2,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "rate(diskio_io_time[1m])/10 > 99",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    },
    {
      "name": "Hard disk - expected to be written full in 4 hours - telegraf",
      "note": "",
      "severity": 1,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "predict_linear(disk_free[1h], 4*3600) < 0",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    },
    {
      "name": "packet loss in the inbound direction - telegraf",
      "note": "",
      "severity": 3,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "increase(net_drop_in[1m]) > 0",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    },
    {
      "name": "packet loss in the outbound direction - telegraf",
      "note": "",
      "severity": 3,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "increase(net_drop_out[1m]) > 0",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    },
    {
      "name": "Network connection - number of TME_WAIT exceeds 20,000 - telegraf",
      "note": "",
      "severity": 2,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "netstat_tcp_time_wait > 20000",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    },
    {
      "name": "there are processes with 0 count, a certain process may have crashed - telegraf",
      "note": "",
      "severity": 1,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "procstat_lookup_running == 0",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    },
    {
      "name": "process handle limit is too small - telegraf",
      "note": "",
      "severity": 3,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "procstat_rlimit_num_fds_soft < 2048",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    },
    {
      "name": "Process monitoring - lookup failure - telegraf",
      "note": "",
      "severity": 1,
      "disabled": 0,
      "prom_for_duration": 60,
      "prom_ql": "procstat_lookup_result_code != 0",
      "prom_eval_interval": 15,
      "enable_stime": "00:00",
      "enable_etime": "23:59",
      "enable_days_of_week": [
        "1",
        "2",
        "3",
        "4",
        "5",
        "6",
        "0"
      ],
      "notify_recovered": 1,
      "notify_channels": [
        "email",
        "dingtalk",
        "wecom"
      ],
      "notify_repeat_step": 60,
      "callbacks": [],
      "runbook_url": "",
      "append_tags": []
    }
  ]