Change alerts
This commit is contained in:
parent
592405b07e
commit
22b07e2da6
1 changed files with 711 additions and 163 deletions
|
|
@ -12,7 +12,7 @@
|
|||
"rps_alert": {
|
||||
"relative_time_range_from": 1800,
|
||||
"keep_firing_for": "",
|
||||
"annotation_summary": "Превышен порог в {rps_limit} RPS в {client_title}",
|
||||
"annotation_summary": "🔴 {client_title} инфраструктурный RPS > {rps_limit} RPS.",
|
||||
"queries": {
|
||||
"data_query": {
|
||||
"refId": "A",
|
||||
|
|
@ -141,7 +141,7 @@
|
|||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "🖥 {{ $labels.instance }}: диск / заполнен на {{ $values.\"Занято в /\".Value | printf \"%.1f\" }}% (порог 75%)"
|
||||
"summary": "🟡 {{ $labels.instance }}: диск / заполнен на {{ $values.\"Занято в /\".Value | printf \"%.1f\" }}% (порог 75%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
|
|
@ -219,161 +219,7 @@
|
|||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "🖥 {{ $labels.instance }}: диск /var/log заполнен на {{ $values.\"Занято в /var/log\".Value | printf \"%.1f\" }}% (порог 75%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
"uid": "ef4spdg2qcge8c",
|
||||
"title": "[SYS] CPU Busy 75%",
|
||||
"condition": "C",
|
||||
"for": "3m",
|
||||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"data": [
|
||||
{
|
||||
"refId": "Занято CPU",
|
||||
"relativeTimeRange": {
|
||||
"from": 600,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "deibaq3pwytxce",
|
||||
"model": {
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "deibaq3pwytxce"
|
||||
},
|
||||
"editorMode": "code",
|
||||
"expr": "100 - (\n avg by(instance) (\n rate(node_cpu_seconds_total{mode=\"idle\", instance=~\"PTAF-TB-.*\"}[5m])\n ) * 100\n)",
|
||||
"instant": true,
|
||||
"intervalMs": 1000,
|
||||
"legendFormat": "__auto",
|
||||
"maxDataPoints": 43200,
|
||||
"range": false,
|
||||
"refId": "Занято CPU"
|
||||
}
|
||||
},
|
||||
{
|
||||
"refId": "C",
|
||||
"relativeTimeRange": {
|
||||
"from": 0,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "__expr__",
|
||||
"model": {
|
||||
"conditions": [
|
||||
{
|
||||
"evaluator": {
|
||||
"params": [
|
||||
75
|
||||
],
|
||||
"type": "gt"
|
||||
},
|
||||
"operator": {
|
||||
"type": "and"
|
||||
},
|
||||
"query": {
|
||||
"params": [
|
||||
"C"
|
||||
]
|
||||
},
|
||||
"reducer": {
|
||||
"params": [],
|
||||
"type": "last"
|
||||
},
|
||||
"type": "query"
|
||||
}
|
||||
],
|
||||
"datasource": {
|
||||
"type": "__expr__",
|
||||
"uid": "__expr__"
|
||||
},
|
||||
"expression": "Занято CPU",
|
||||
"intervalMs": 1000,
|
||||
"maxDataPoints": 43200,
|
||||
"refId": "C",
|
||||
"type": "threshold"
|
||||
}
|
||||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "🔥 {{ $labels.instance }}: CPU загружен на {{ $values.\"Занято CPU\".Value | printf \"%.1f\" }}% (порог 75%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
"uid": "ef4spmgmowe80e",
|
||||
"title": "[SYS] RAM Busy 75%",
|
||||
"condition": "Превышение потребления RAM",
|
||||
"for": "3m",
|
||||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"data": [
|
||||
{
|
||||
"refId": "Занято RAM",
|
||||
"relativeTimeRange": {
|
||||
"from": 600,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "deibaq3pwytxce",
|
||||
"model": {
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "deibaq3pwytxce"
|
||||
},
|
||||
"editorMode": "code",
|
||||
"expr": "100 - (\n (node_memory_MemAvailable_bytes{instance=~\"PTAF-TB-.*\"} / \n node_memory_MemTotal_bytes{instance=~\"PTAF-TB-.*\"}) * 100\n)",
|
||||
"instant": true,
|
||||
"intervalMs": 1000,
|
||||
"legendFormat": "__auto",
|
||||
"maxDataPoints": 43200,
|
||||
"range": false,
|
||||
"refId": "Занято RAM"
|
||||
}
|
||||
},
|
||||
{
|
||||
"refId": "Превышение потребления RAM",
|
||||
"relativeTimeRange": {
|
||||
"from": 0,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "__expr__",
|
||||
"model": {
|
||||
"conditions": [
|
||||
{
|
||||
"evaluator": {
|
||||
"params": [
|
||||
75
|
||||
],
|
||||
"type": "gt"
|
||||
},
|
||||
"operator": {
|
||||
"type": "and"
|
||||
},
|
||||
"query": {
|
||||
"params": [
|
||||
"C"
|
||||
]
|
||||
},
|
||||
"reducer": {
|
||||
"params": [],
|
||||
"type": "last"
|
||||
},
|
||||
"type": "query"
|
||||
}
|
||||
],
|
||||
"datasource": {
|
||||
"type": "__expr__",
|
||||
"uid": "__expr__"
|
||||
},
|
||||
"expression": "Занято RAM",
|
||||
"intervalMs": 1000,
|
||||
"maxDataPoints": 43200,
|
||||
"refId": "Превышение потребления RAM",
|
||||
"type": "threshold"
|
||||
}
|
||||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "💾 {{ $labels.instance }}: RAM занята на {{ $values.\"Занято RAM\".Value | printf \"%.1f\" }}% (порог 75%)"
|
||||
"summary": "🟡 {{ $labels.instance }}: диск /var/log заполнен на {{ $values.\"Занято в /var/log\".Value | printf \"%.1f\" }}% (порог 75%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
|
|
@ -451,7 +297,7 @@
|
|||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "🚨 {{ $labels.instance }}: диск / заполнен на {{ $values.\"Занято в /\".Value | printf \"%.1f\" }}% (порог 90%)"
|
||||
"summary": "🔴 {{ $labels.instance }}: диск / заполнен на {{ $values.\"Занято в /\".Value | printf \"%.1f\" }}% (порог 90%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
|
|
@ -529,7 +375,469 @@
|
|||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "🚨 {{ $labels.instance }}: диск /var/log заполнен на {{ $values.\"Занято в /var/log\".Value | printf \"%.1f\" }}% (порог 90%)"
|
||||
"summary": "🔴 {{ $labels.instance }}: диск /var/log заполнен на {{ $values.\"Занято в /var/log\".Value | printf \"%.1f\" }}% (порог 90%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
"uid": "ef4spdg2qcge35",
|
||||
"title": "[SYS] CPU Busy 35%",
|
||||
"condition": "C",
|
||||
"for": "5m",
|
||||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"data": [
|
||||
{
|
||||
"refId": "Занято CPU",
|
||||
"relativeTimeRange": {
|
||||
"from": 600,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "deibaq3pwytxce",
|
||||
"model": {
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "deibaq3pwytxce"
|
||||
},
|
||||
"editorMode": "code",
|
||||
"expr": "100 - (\n avg by(instance) (\n rate(node_cpu_seconds_total{mode=\"idle\", instance=~\"PTAF-TB-.*\"}[5m])\n ) * 100\n)",
|
||||
"instant": true,
|
||||
"intervalMs": 1000,
|
||||
"legendFormat": "__auto",
|
||||
"maxDataPoints": 43200,
|
||||
"range": false,
|
||||
"refId": "Занято CPU"
|
||||
}
|
||||
},
|
||||
{
|
||||
"refId": "C",
|
||||
"relativeTimeRange": {
|
||||
"from": 0,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "__expr__",
|
||||
"model": {
|
||||
"conditions": [
|
||||
{
|
||||
"evaluator": {
|
||||
"params": [
|
||||
35
|
||||
],
|
||||
"type": "gt"
|
||||
},
|
||||
"operator": {
|
||||
"type": "and"
|
||||
},
|
||||
"query": {
|
||||
"params": [
|
||||
"C"
|
||||
]
|
||||
},
|
||||
"reducer": {
|
||||
"params": [],
|
||||
"type": "last"
|
||||
},
|
||||
"type": "query"
|
||||
}
|
||||
],
|
||||
"datasource": {
|
||||
"type": "__expr__",
|
||||
"uid": "__expr__"
|
||||
},
|
||||
"expression": "Занято CPU",
|
||||
"intervalMs": 1000,
|
||||
"maxDataPoints": 43200,
|
||||
"refId": "C",
|
||||
"type": "threshold"
|
||||
}
|
||||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "⚪ {{ $labels.instance }}: CPU загружен на {{ $values.\"Занято CPU\".Value | printf \"%.1f\" }}% (порог 35%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
"uid": "ef4spdg2qcge45",
|
||||
"title": "[SYS] CPU Busy 45%",
|
||||
"condition": "C",
|
||||
"for": "5m",
|
||||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"data": [
|
||||
{
|
||||
"refId": "Занято CPU",
|
||||
"relativeTimeRange": {
|
||||
"from": 600,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "deibaq3pwytxce",
|
||||
"model": {
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "deibaq3pwytxce"
|
||||
},
|
||||
"editorMode": "code",
|
||||
"expr": "100 - (\n avg by(instance) (\n rate(node_cpu_seconds_total{mode=\"idle\", instance=~\"PTAF-TB-.*\"}[5m])\n ) * 100\n)",
|
||||
"instant": true,
|
||||
"intervalMs": 1000,
|
||||
"legendFormat": "__auto",
|
||||
"maxDataPoints": 43200,
|
||||
"range": false,
|
||||
"refId": "Занято CPU"
|
||||
}
|
||||
},
|
||||
{
|
||||
"refId": "C",
|
||||
"relativeTimeRange": {
|
||||
"from": 0,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "__expr__",
|
||||
"model": {
|
||||
"conditions": [
|
||||
{
|
||||
"evaluator": {
|
||||
"params": [
|
||||
45
|
||||
],
|
||||
"type": "gt"
|
||||
},
|
||||
"operator": {
|
||||
"type": "and"
|
||||
},
|
||||
"query": {
|
||||
"params": [
|
||||
"C"
|
||||
]
|
||||
},
|
||||
"reducer": {
|
||||
"params": [],
|
||||
"type": "last"
|
||||
},
|
||||
"type": "query"
|
||||
}
|
||||
],
|
||||
"datasource": {
|
||||
"type": "__expr__",
|
||||
"uid": "__expr__"
|
||||
},
|
||||
"expression": "Занято CPU",
|
||||
"intervalMs": 1000,
|
||||
"maxDataPoints": 43200,
|
||||
"refId": "C",
|
||||
"type": "threshold"
|
||||
}
|
||||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "🟡 {{ $labels.instance }}: CPU загружен на {{ $values.\"Занято CPU\".Value | printf \"%.1f\" }}% (порог 45%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
"uid": "ef4spdg2qcge8c",
|
||||
"title": "[SYS] CPU Busy 75%",
|
||||
"condition": "C",
|
||||
"for": "3m",
|
||||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"data": [
|
||||
{
|
||||
"refId": "Занято CPU",
|
||||
"relativeTimeRange": {
|
||||
"from": 600,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "deibaq3pwytxce",
|
||||
"model": {
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "deibaq3pwytxce"
|
||||
},
|
||||
"editorMode": "code",
|
||||
"expr": "100 - (\n avg by(instance) (\n rate(node_cpu_seconds_total{mode=\"idle\", instance=~\"PTAF-TB-.*\"}[5m])\n ) * 100\n)",
|
||||
"instant": true,
|
||||
"intervalMs": 1000,
|
||||
"legendFormat": "__auto",
|
||||
"maxDataPoints": 43200,
|
||||
"range": false,
|
||||
"refId": "Занято CPU"
|
||||
}
|
||||
},
|
||||
{
|
||||
"refId": "C",
|
||||
"relativeTimeRange": {
|
||||
"from": 0,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "__expr__",
|
||||
"model": {
|
||||
"conditions": [
|
||||
{
|
||||
"evaluator": {
|
||||
"params": [
|
||||
75
|
||||
],
|
||||
"type": "gt"
|
||||
},
|
||||
"operator": {
|
||||
"type": "and"
|
||||
},
|
||||
"query": {
|
||||
"params": [
|
||||
"C"
|
||||
]
|
||||
},
|
||||
"reducer": {
|
||||
"params": [],
|
||||
"type": "last"
|
||||
},
|
||||
"type": "query"
|
||||
}
|
||||
],
|
||||
"datasource": {
|
||||
"type": "__expr__",
|
||||
"uid": "__expr__"
|
||||
},
|
||||
"expression": "Занято CPU",
|
||||
"intervalMs": 1000,
|
||||
"maxDataPoints": 43200,
|
||||
"refId": "C",
|
||||
"type": "threshold"
|
||||
}
|
||||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "🔴 {{ $labels.instance }}: CPU загружен на {{ $values.\"Занято CPU\".Value | printf \"%.1f\" }}% (порог 75%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
"uid": "ef4spmgmowe35",
|
||||
"title": "[SYS] RAM Busy 35%",
|
||||
"condition": "Превышение потребления RAM",
|
||||
"for": "5m",
|
||||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"data": [
|
||||
{
|
||||
"refId": "Занято RAM",
|
||||
"relativeTimeRange": {
|
||||
"from": 600,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "deibaq3pwytxce",
|
||||
"model": {
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "deibaq3pwytxce"
|
||||
},
|
||||
"editorMode": "code",
|
||||
"expr": "100 - (\n (node_memory_MemAvailable_bytes{instance=~\"PTAF-TB-.*\"} / \n node_memory_MemTotal_bytes{instance=~\"PTAF-TB-.*\"}) * 100\n)",
|
||||
"instant": true,
|
||||
"intervalMs": 1000,
|
||||
"legendFormat": "__auto",
|
||||
"maxDataPoints": 43200,
|
||||
"range": false,
|
||||
"refId": "Занято RAM"
|
||||
}
|
||||
},
|
||||
{
|
||||
"refId": "Превышение потребления RAM",
|
||||
"relativeTimeRange": {
|
||||
"from": 0,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "__expr__",
|
||||
"model": {
|
||||
"conditions": [
|
||||
{
|
||||
"evaluator": {
|
||||
"params": [
|
||||
35
|
||||
],
|
||||
"type": "gt"
|
||||
},
|
||||
"operator": {
|
||||
"type": "and"
|
||||
},
|
||||
"query": {
|
||||
"params": [
|
||||
"C"
|
||||
]
|
||||
},
|
||||
"reducer": {
|
||||
"params": [],
|
||||
"type": "last"
|
||||
},
|
||||
"type": "query"
|
||||
}
|
||||
],
|
||||
"datasource": {
|
||||
"type": "__expr__",
|
||||
"uid": "__expr__"
|
||||
},
|
||||
"expression": "Занято RAM",
|
||||
"intervalMs": 1000,
|
||||
"maxDataPoints": 43200,
|
||||
"refId": "Превышение потребления RAM",
|
||||
"type": "threshold"
|
||||
}
|
||||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "⚪ {{ $labels.instance }}: RAM занята на {{ $values.\"Занято RAM\".Value | printf \"%.1f\" }}% (порог 35%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
"uid": "ef4spmgmowe45",
|
||||
"title": "[SYS] RAM Busy 45%",
|
||||
"condition": "Превышение потребления RAM",
|
||||
"for": "5m",
|
||||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"data": [
|
||||
{
|
||||
"refId": "Занято RAM",
|
||||
"relativeTimeRange": {
|
||||
"from": 600,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "deibaq3pwytxce",
|
||||
"model": {
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "deibaq3pwytxce"
|
||||
},
|
||||
"editorMode": "code",
|
||||
"expr": "100 - (\n (node_memory_MemAvailable_bytes{instance=~\"PTAF-TB-.*\"} / \n node_memory_MemTotal_bytes{instance=~\"PTAF-TB-.*\"}) * 100\n)",
|
||||
"instant": true,
|
||||
"intervalMs": 1000,
|
||||
"legendFormat": "__auto",
|
||||
"maxDataPoints": 43200,
|
||||
"range": false,
|
||||
"refId": "Занято RAM"
|
||||
}
|
||||
},
|
||||
{
|
||||
"refId": "Превышение потребления RAM",
|
||||
"relativeTimeRange": {
|
||||
"from": 0,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "__expr__",
|
||||
"model": {
|
||||
"conditions": [
|
||||
{
|
||||
"evaluator": {
|
||||
"params": [
|
||||
45
|
||||
],
|
||||
"type": "gt"
|
||||
},
|
||||
"operator": {
|
||||
"type": "and"
|
||||
},
|
||||
"query": {
|
||||
"params": [
|
||||
"C"
|
||||
]
|
||||
},
|
||||
"reducer": {
|
||||
"params": [],
|
||||
"type": "last"
|
||||
},
|
||||
"type": "query"
|
||||
}
|
||||
],
|
||||
"datasource": {
|
||||
"type": "__expr__",
|
||||
"uid": "__expr__"
|
||||
},
|
||||
"expression": "Занято RAM",
|
||||
"intervalMs": 1000,
|
||||
"maxDataPoints": 43200,
|
||||
"refId": "Превышение потребления RAM",
|
||||
"type": "threshold"
|
||||
}
|
||||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "🟡 {{ $labels.instance }}: RAM занята на {{ $values.\"Занято RAM\".Value | printf \"%.1f\" }}% (порог 45%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
"uid": "ef4spmgmowe80e",
|
||||
"title": "[SYS] RAM Busy 75%",
|
||||
"condition": "Превышение потребления RAM",
|
||||
"for": "3m",
|
||||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"data": [
|
||||
{
|
||||
"refId": "Занято RAM",
|
||||
"relativeTimeRange": {
|
||||
"from": 600,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "deibaq3pwytxce",
|
||||
"model": {
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "deibaq3pwytxce"
|
||||
},
|
||||
"editorMode": "code",
|
||||
"expr": "100 - (\n (node_memory_MemAvailable_bytes{instance=~\"PTAF-TB-.*\"} / \n node_memory_MemTotal_bytes{instance=~\"PTAF-TB-.*\"}) * 100\n)",
|
||||
"instant": true,
|
||||
"intervalMs": 1000,
|
||||
"legendFormat": "__auto",
|
||||
"maxDataPoints": 43200,
|
||||
"range": false,
|
||||
"refId": "Занято RAM"
|
||||
}
|
||||
},
|
||||
{
|
||||
"refId": "Превышение потребления RAM",
|
||||
"relativeTimeRange": {
|
||||
"from": 0,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "__expr__",
|
||||
"model": {
|
||||
"conditions": [
|
||||
{
|
||||
"evaluator": {
|
||||
"params": [
|
||||
75
|
||||
],
|
||||
"type": "gt"
|
||||
},
|
||||
"operator": {
|
||||
"type": "and"
|
||||
},
|
||||
"query": {
|
||||
"params": [
|
||||
"C"
|
||||
]
|
||||
},
|
||||
"reducer": {
|
||||
"params": [],
|
||||
"type": "last"
|
||||
},
|
||||
"type": "query"
|
||||
}
|
||||
],
|
||||
"datasource": {
|
||||
"type": "__expr__",
|
||||
"uid": "__expr__"
|
||||
},
|
||||
"expression": "Занято RAM",
|
||||
"intervalMs": 1000,
|
||||
"maxDataPoints": 43200,
|
||||
"refId": "Превышение потребления RAM",
|
||||
"type": "threshold"
|
||||
}
|
||||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "🔴 {{ $labels.instance }}: RAM занята на {{ $values.\"Занято RAM\".Value | printf \"%.1f\" }}% (порог 75%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
|
|
@ -606,7 +914,7 @@
|
|||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "🚨 {{ $labels.instance }}: CPU загружен на {{ $values.\"Занято CPU\".Value | printf \"%.1f\" }}% (порог 90%)"
|
||||
"summary": "🔴 {{ $labels.instance }}: CPU загружен на {{ $values.\"Занято CPU\".Value | printf \"%.1f\" }}% (порог 90%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
|
|
@ -683,9 +991,249 @@
|
|||
}
|
||||
],
|
||||
"annotations": {
|
||||
"summary": "🚨 {{ $labels.instance }}: RAM занята на {{ $values.\"Занято RAM\".Value | printf \"%.1f\" }}% (порог 90%)"
|
||||
"summary": "🔴 {{ $labels.instance }}: RAM занята на {{ $values.\"Занято RAM\".Value | printf \"%.1f\" }}% (порог 90%)"
|
||||
}
|
||||
},
|
||||
{
|
||||
"uid": "net_total_35_bond0",
|
||||
"title": "[SYS] Network bond0 35%",
|
||||
"condition": "Превышение",
|
||||
"for": "5m",
|
||||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"isPaused": false,
|
||||
"folderUID": "",
|
||||
"ruleGroup": "",
|
||||
"annotations": {
|
||||
"summary": "⚪ {{ $labels.instance }}: нагрузка на bond0 {{ $values.\"Total\".Value | printf \"%.2f\" }} Гбит/с (порог 35% от 25 Гбит/с)"
|
||||
},
|
||||
"data": [
|
||||
{
|
||||
"refId": "Total",
|
||||
"relativeTimeRange": {
|
||||
"from": 600,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "deibaq3pwytxce",
|
||||
"model": {
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "deibaq3pwytxce"
|
||||
},
|
||||
"editorMode": "code",
|
||||
"expr": "(rate(node_network_receive_bytes_total{instance=~\"PTAF-TB-.*\", device=\"bond0\"}[5m]) + rate(node_network_transmit_bytes_total{instance=~\"PTAF-TB-.*\", device=\"bond0\"}[5m])) * 8 / 1000000000",
|
||||
"instant": true,
|
||||
"intervalMs": 1000,
|
||||
"legendFormat": "__auto",
|
||||
"maxDataPoints": 43200,
|
||||
"range": false,
|
||||
"refId": "Total"
|
||||
}
|
||||
},
|
||||
{
|
||||
"refId": "Превышение",
|
||||
"relativeTimeRange": {
|
||||
"from": 0,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "__expr__",
|
||||
"model": {
|
||||
"conditions": [
|
||||
{
|
||||
"evaluator": {
|
||||
"params": [
|
||||
8.75
|
||||
],
|
||||
"type": "gt"
|
||||
},
|
||||
"operator": {
|
||||
"type": "and"
|
||||
},
|
||||
"query": {
|
||||
"params": [
|
||||
"Превышение"
|
||||
]
|
||||
},
|
||||
"reducer": {
|
||||
"params": [],
|
||||
"type": "last"
|
||||
},
|
||||
"type": "query"
|
||||
}
|
||||
],
|
||||
"datasource": {
|
||||
"type": "__expr__",
|
||||
"uid": "__expr__"
|
||||
},
|
||||
"expression": "Total",
|
||||
"intervalMs": 1000,
|
||||
"maxDataPoints": 43200,
|
||||
"refId": "Превышение",
|
||||
"type": "threshold"
|
||||
}
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"uid": "net_total_45_bond0",
|
||||
"title": "[SYS] Network bond0 45%",
|
||||
"condition": "Превышение",
|
||||
"for": "3m",
|
||||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"isPaused": false,
|
||||
"folderUID": "",
|
||||
"ruleGroup": "",
|
||||
"annotations": {
|
||||
"summary": "🟡 {{ $labels.instance }}: нагрузка на bond0 {{ $values.\"Total\".Value | printf \"%.2f\" }} Гбит/с (порог 45% от 25 Гбит/с)"
|
||||
},
|
||||
"data": [
|
||||
{
|
||||
"refId": "Total",
|
||||
"relativeTimeRange": {
|
||||
"from": 600,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "deibaq3pwytxce",
|
||||
"model": {
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "deibaq3pwytxce"
|
||||
},
|
||||
"editorMode": "code",
|
||||
"expr": "(rate(node_network_receive_bytes_total{instance=~\"PTAF-TB-.*\", device=\"bond0\"}[5m]) + rate(node_network_transmit_bytes_total{instance=~\"PTAF-TB-.*\", device=\"bond0\"}[5m])) * 8 / 1000000000",
|
||||
"instant": true,
|
||||
"intervalMs": 1000,
|
||||
"legendFormat": "__auto",
|
||||
"maxDataPoints": 43200,
|
||||
"range": false,
|
||||
"refId": "Total"
|
||||
}
|
||||
},
|
||||
{
|
||||
"refId": "Превышение",
|
||||
"relativeTimeRange": {
|
||||
"from": 0,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "__expr__",
|
||||
"model": {
|
||||
"conditions": [
|
||||
{
|
||||
"evaluator": {
|
||||
"params": [
|
||||
11.25
|
||||
],
|
||||
"type": "gt"
|
||||
},
|
||||
"operator": {
|
||||
"type": "and"
|
||||
},
|
||||
"query": {
|
||||
"params": [
|
||||
"Превышение"
|
||||
]
|
||||
},
|
||||
"reducer": {
|
||||
"params": [],
|
||||
"type": "last"
|
||||
},
|
||||
"type": "query"
|
||||
}
|
||||
],
|
||||
"datasource": {
|
||||
"type": "__expr__",
|
||||
"uid": "__expr__"
|
||||
},
|
||||
"expression": "Total",
|
||||
"intervalMs": 1000,
|
||||
"maxDataPoints": 43200,
|
||||
"refId": "Превышение",
|
||||
"type": "threshold"
|
||||
}
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"uid": "net_total_75_bond0",
|
||||
"title": "[SYS] Network bond0 75%",
|
||||
"condition": "Превышение",
|
||||
"for": "1m",
|
||||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"isPaused": false,
|
||||
"folderUID": "",
|
||||
"ruleGroup": "",
|
||||
"annotations": {
|
||||
"summary": "🔴 {{ $labels.instance }}: нагрузка на bond0 {{ $values.\"Total\".Value | printf \"%.2f\" }} Гбит/с (порог 75% от 25 Гбит/с)"
|
||||
},
|
||||
"data": [
|
||||
{
|
||||
"refId": "Total",
|
||||
"relativeTimeRange": {
|
||||
"from": 600,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "deibaq3pwytxce",
|
||||
"model": {
|
||||
"datasource": {
|
||||
"type": "prometheus",
|
||||
"uid": "deibaq3pwytxce"
|
||||
},
|
||||
"editorMode": "code",
|
||||
"expr": "(rate(node_network_receive_bytes_total{instance=~\"PTAF-TB-.*\", device=\"bond0\"}[5m]) + rate(node_network_transmit_bytes_total{instance=~\"PTAF-TB-.*\", device=\"bond0\"}[5m])) * 8 / 1000000000",
|
||||
"instant": true,
|
||||
"intervalMs": 1000,
|
||||
"legendFormat": "__auto",
|
||||
"maxDataPoints": 43200,
|
||||
"range": false,
|
||||
"refId": "Total"
|
||||
}
|
||||
},
|
||||
{
|
||||
"refId": "Превышение",
|
||||
"relativeTimeRange": {
|
||||
"from": 0,
|
||||
"to": 0
|
||||
},
|
||||
"datasourceUid": "__expr__",
|
||||
"model": {
|
||||
"conditions": [
|
||||
{
|
||||
"evaluator": {
|
||||
"params": [
|
||||
18.75
|
||||
],
|
||||
"type": "gt"
|
||||
},
|
||||
"operator": {
|
||||
"type": "and"
|
||||
},
|
||||
"query": {
|
||||
"params": [
|
||||
"Превышение"
|
||||
]
|
||||
},
|
||||
"reducer": {
|
||||
"params": [],
|
||||
"type": "last"
|
||||
},
|
||||
"type": "query"
|
||||
}
|
||||
],
|
||||
"datasource": {
|
||||
"type": "__expr__",
|
||||
"uid": "__expr__"
|
||||
},
|
||||
"expression": "Total",
|
||||
"intervalMs": 1000,
|
||||
"maxDataPoints": 43200,
|
||||
"refId": "Превышение",
|
||||
"type": "threshold"
|
||||
}
|
||||
}
|
||||
]
|
||||
},
|
||||
{
|
||||
"uid": "cfc4iwprdvn5se",
|
||||
"title": "[SYS] Состояние контейнеров",
|
||||
|
|
@ -694,7 +1242,7 @@
|
|||
"no_data_state": "NoData",
|
||||
"exec_err_state": "Error",
|
||||
"annotations": {
|
||||
"summary": "🐳 Контейнер {{ index $labels `name` }} на хосте {{ index $labels `instance` }} упал. Текущее значение: {{ $values.A.Value }}"
|
||||
"summary": "🔴 Контейнер {{ index $labels `name` }} на хосте {{ index $labels `instance` }} упал. Текущее значение: {{ $values.A.Value }}"
|
||||
},
|
||||
"data": [
|
||||
{
|
||||
|
|
@ -844,6 +1392,6 @@
|
|||
],
|
||||
"contact_point": {
|
||||
"receiver_name": "tg+mail+mattermost PTAF Grafana",
|
||||
"message_template": "{{ range .Alerts }}PTAF alerts: \n{{ if eq .Annotations.severity \"warning\" }}🟡{{ else }}🔴{{ end }} {{ if .Annotations.client_domain }}{{ .Annotations.client_domain }}{{ end }}{{ if .Annotations.code_field }} — {{ .Annotations.code_field }} ошибок{{ end }}{{ if eq .Status \"firing\" }}{{ if index .Values \"Значение\" }} {{ index .Values \"Значение\" | printf \"%.2f\" }}{{ if .Annotations.value_unit }} {{ .Annotations.value_unit }}{{ else }}%{{ end }}{{ end }}{{ if index .Values \"_Reduce\" }} {{ index .Values \"_Reduce\" | printf \"%.0f\" }}{{ if .Annotations.value_unit }} {{ .Annotations.value_unit }}{{ else }}%{{ end }}{{ end }} за {{ .Annotations.for_duration }}.{{ else }} трафика.{{ end }}\n{{ if eq .Status \"firing\" }}{{ if .Annotations.threshold }}Порог сработки {{ .Annotations.threshold }}\n{{ end }}{{ else }}{{ if index .Values \"Значение\" }}Текущее значение {{ index .Values \"Значение\" | printf \"%.2f\" }}{{ if .Annotations.value_unit }} {{ .Annotations.value_unit }}{{ else }}%{{ end }} {{ if .Annotations.field_info }}{{ .Annotations.field_info }}{{ end }}\n{{ end }}{{ if index .Values \"_Reduce\" }}Текущее значение {{ index .Values \"_Reduce\" | printf \"%.0f\" }}{{ if .Annotations.value_unit }} {{ .Annotations.value_unit }}{{ else }}%{{ end }} {{ if .Annotations.field_info }}{{ .Annotations.field_info }}{{ end }}\n{{ end }}{{ end }}Время: {{ .StartsAt.Format \"2006-01-02 15:04:05\" }}\n{{ if .Annotations.runbook }}---\n{{ .Annotations.runbook }}\n{{ end }}{{ end }}"
|
||||
"message_template": "{{ range .Alerts }}{{ if eq .Status \"firing\" }}🔴 Firing{{ else }}✅ Resolved{{ end }}\n{{ if .Annotations.client_domain }}{{ if eq .Annotations.severity \"warning\" }}🟡 {{ .Annotations.client_domain }}{{ if index .Values \"_Reduce\" }} — на WAF {{ index .Values \"_Reduce\" | printf \"%.0f\" }} {{ if .Annotations.value_unit }}{{ .Annotations.value_unit }}{{ end }}{{ end }} {{ if .Annotations.code_field }}{{ .Annotations.code_field }}{{ end }} кодов ответа {{ if .Annotations.field_info }}{{ .Annotations.field_info }}{{ end }}\n{{ if .Annotations.threshold_count }}Пороговое значение {{ .Annotations.threshold_count }}\n{{ end }}{{ else }}{{ if eq .Status \"firing\" }}🔴 {{ .Annotations.client_domain }} — {{ .Annotations.code_field }} ошибок > {{ .Annotations.threshold_value }}% держится > {{ .Annotations.for_duration }}.\nТекущее значение: {{ if index .Values \"Значение\" }}{{ index .Values \"Значение\" | printf \"%.2f\" }}%{{ end }}{{ if index .Values \"_Reduce\" }}{{ index .Values \"_Reduce\" | printf \"%.2f\" }}%{{ end }} {{ if .Annotations.field_info }}{{ .Annotations.field_info }}{{ end }}\n{{ else }}🔴 {{ .Annotations.client_domain }} — {{ .Annotations.code_field }} ошибок < {{ .Annotations.threshold_value }}%.\nТекущее значение: {{ if index .Values \"Значение\" }}{{ index .Values \"Значение\" | printf \"%.2f\" }}%{{ end }}{{ if index .Values \"_Reduce\" }}{{ index .Values \"_Reduce\" | printf \"%.2f\" }}%{{ end }} {{ if .Annotations.field_info }}{{ .Annotations.field_info }}{{ end }}\n{{ end }}{{ end }}{{ else }}{{ if .Annotations.summary }}{{ .Annotations.summary }}\n{{ end }}{{ end }}{{ if eq .Status \"firing\" }}Время сработки алерта: {{ .StartsAt.Format \"2006-01-02 15:04:05 UTC\" }}{{ else }}Время разрешения алерта: {{ .EndsAt.Format \"2006-01-02 15:04:05 UTC\" }}{{ end }}\n{{ if .Annotations.runbook }}\nНеобходимо {{ .Annotations.runbook }}\n{{ end }}{{ end }}"
|
||||
}
|
||||
}
|
||||
Loading…
Add table
Reference in a new issue