nightingale/integrations/Linux/alerts/linux_by_telegraf.json

[
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "Address is unreachable by PING, please pay attention - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 1,
        "severities": [
            1
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "ping_result_code != 0",
                    "severity": 1
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327714548000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "Hard disk - expected to be written full in 4 hours - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 1,
        "severities": [
            1
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "predict_linear(disk_free[1h], 4*3600) \u003c 0",
                    "severity": 1
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327715253000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "Hard disk - IO is very busy - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 2,
        "severities": [
            2
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "rate(diskio_io_time[1m])/10 \u003e 99",
                    "severity": 2
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327715994000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "host",
        "datasource_ids": null,
        "cluster": "",
        "name": "Lost connection with monitoring target - telegraf",
        "note": "",
        "prod": "host",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 0,
        "severities": [
            2
        ],
        "disabled": 1,
        "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_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": {},
        "extra_config": {
            "escalation": {},
            "notify_aggregation": {}
        },
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327717532000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "Machine load - high CPU, please pay attention - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 3,
        "severities": [
            3
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "cpu_usage_idle{cpu=\"cpu-total\"} \u003c 25",
                    "severity": 3
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327718353000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "Machine load - high memory, please pay attention - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 2,
        "severities": [
            2
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "mem_available_percent \u003c 25",
                    "severity": 2
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327719248000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "Network connection - number of TME_WAIT exceeds 20,000 - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 2,
        "severities": [
            2
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "netstat_tcp_time_wait \u003e 20000",
                    "severity": 2
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327720659000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "packet loss in the inbound direction - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 3,
        "severities": [
            3
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "increase(net_drop_in[1m]) \u003e 0",
                    "severity": 3
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327721904000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "packet loss in the outbound direction - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 3,
        "severities": [
            3
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "increase(net_drop_out[1m]) \u003e 0",
                    "severity": 3
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327722565000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "Port detection failed, please pay attention - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 1,
        "severities": [
            1
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "net_response_result_code != 0",
                    "severity": 1
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327723152000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "process handle limit is too small - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 3,
        "severities": [
            3
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "procstat_rlimit_num_fds_soft \u003c 2048",
                    "severity": 3
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327723625000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "Process monitoring - lookup failure - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 1,
        "severities": [
            1
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "procstat_lookup_result_code != 0",
                    "severity": 1
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327724170000
    },
    {
        "id": 0,
        "group_id": 0,
        "cate": "prometheus",
        "datasource_ids": [
            0
        ],
        "cluster": "",
        "name": "there are processes with 0 count, a certain process may have crashed - telegraf",
        "note": "",
        "prod": "metric",
        "algorithm": "",
        "algo_params": null,
        "delay": 0,
        "severity": 1,
        "severities": [
            1
        ],
        "disabled": 1,
        "prom_for_duration": 60,
        "prom_ql": "",
        "rule_config": {
            "algo_params": null,
            "inhibit": false,
            "prom_ql": "",
            "queries": [
                {
                    "prom_ql": "procstat_lookup_running == 0",
                    "severity": 1
                }
            ],
            "severity": 0
        },
        "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": [
            "email",
            "dingtalk",
            "wecom"
        ],
        "notify_groups_obj": null,
        "notify_groups": null,
        "notify_repeat_step": 60,
        "notify_max_number": 0,
        "recover_duration": 0,
        "callbacks": [],
        "runbook_url": "",
        "append_tags": [],
        "annotations": null,
        "extra_config": null,
        "create_at": 0,
        "create_by": "",
        "update_at": 0,
        "update_by": "",
        "uuid": 1717556327724727000
    }
]