752 lines
13 KiB
Plaintext
752 lines
13 KiB
Plaintext
[
|
|
{
|
|
"name": "timewait状态tcp连接超过2万",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 3,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "net.sockets.tcp.timewait",
|
|
"params": [],
|
|
"threshold": 20000
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": [],
|
|
"notify_user": [],
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "内存利用率大于75%",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 2,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "mem.bytes.used.percent",
|
|
"params": [],
|
|
"threshold": 75
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "机器loadavg大于16",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 2,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "cpu.loadavg.1",
|
|
"params": [],
|
|
"threshold": 16
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "某磁盘无法正常读写",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 1,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "disk.rw.error",
|
|
"params": [],
|
|
"threshold": 0
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "监控agent失联",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 1,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": "=",
|
|
"func": "nodata",
|
|
"metric": "proc.agent.alive",
|
|
"params": [],
|
|
"threshold": 0
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "磁盘利用率达到85%",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 3,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "disk.bytes.used.percent",
|
|
"params": [],
|
|
"threshold": 85
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "磁盘利用率达到88%",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 2,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "disk.bytes.used.percent",
|
|
"params": [],
|
|
"threshold": 88
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "磁盘利用率达到92%",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 1,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "disk.bytes.used.percent",
|
|
"params": [],
|
|
"threshold": 92
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "端口挂了",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 2,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": "!=",
|
|
"func": "all",
|
|
"metric": "proc.port.listen",
|
|
"params": [],
|
|
"threshold": 1
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "网卡入方向丢包",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 2,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "net.in.dropped",
|
|
"params": [],
|
|
"threshold": 3
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "网卡入方向错包",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 2,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "net.in.errs",
|
|
"params": [],
|
|
"threshold": 3
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "网卡出方向丢包",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 2,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "net.out.dropped",
|
|
"params": [],
|
|
"threshold": 3
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "网卡出方向错包",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 2,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "net.out.errs",
|
|
"params": [],
|
|
"threshold": 3
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "进程总数超过3000",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 1,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": ">",
|
|
"func": "all",
|
|
"metric": "sys.ps.process.total",
|
|
"params": [],
|
|
"threshold": 3000
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
},
|
|
{
|
|
"name": "进程挂了",
|
|
"category": 1,
|
|
"alert_dur": 60,
|
|
"recovery_dur": 0,
|
|
"recovery_notify": 1,
|
|
"enable_stime": "00:00",
|
|
"enable_etime": "23:59",
|
|
"priority": 2,
|
|
"callback": "",
|
|
"need_upgrade": 0,
|
|
"runbook": "",
|
|
"excl_nid": null,
|
|
"nids": null,
|
|
"exprs": [
|
|
{
|
|
"eopt": "<",
|
|
"func": "all",
|
|
"metric": "proc.num",
|
|
"params": [],
|
|
"threshold": 1
|
|
}
|
|
],
|
|
"tags": [],
|
|
"enable_days_of_week": [
|
|
0,
|
|
1,
|
|
2,
|
|
3,
|
|
4,
|
|
5,
|
|
6
|
|
],
|
|
"converge": [
|
|
36000,
|
|
1
|
|
],
|
|
"notify_group": null,
|
|
"notify_user": null,
|
|
"leaf_nids": null,
|
|
"endpoints": null,
|
|
"alert_upgrade": {
|
|
"users": [],
|
|
"groups": [],
|
|
"duration": 60,
|
|
"level": 1
|
|
},
|
|
"judge_instance": "",
|
|
"work_groups": null
|
|
}
|
|
] |