forked from flashcat/categraf
642 lines
23 KiB
JSON
642 lines
23 KiB
JSON
{
|
||
"name": "Kubernetes / API Server",
|
||
"tags": "Categraf",
|
||
"configs": {
|
||
"version": "2.0.0",
|
||
"links": [],
|
||
"var": [],
|
||
"panels": [
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "up{job=\"apiserver\"}",
|
||
"legend": "{{ instance }}"
|
||
}
|
||
],
|
||
"name": "API Server - Health Status",
|
||
"links": [],
|
||
"description": "apiserver的实例健康状态，0表示down，1表示up",
|
||
"custom": {
|
||
"showHeader": true,
|
||
"colorMode": "value",
|
||
"calc": "lastNotNull",
|
||
"displayMode": "seriesToRows"
|
||
},
|
||
"options": {
|
||
"valueMappings": [
|
||
{
|
||
"options": {
|
||
"0": {
|
||
"text": "DOWN"
|
||
},
|
||
"1": {
|
||
"text": "UP"
|
||
}
|
||
},
|
||
"type": "value"
|
||
},
|
||
{
|
||
"type": "special",
|
||
"result": {
|
||
"color": "#3fc453",
|
||
"text": "UP"
|
||
},
|
||
"match": {
|
||
"special": 1
|
||
}
|
||
},
|
||
{
|
||
"type": "special",
|
||
"result": {
|
||
"color": "#f80202",
|
||
"text": "DOWN"
|
||
},
|
||
"match": {
|
||
"special": 0
|
||
}
|
||
}
|
||
],
|
||
"standardOptions": {
|
||
"util": "none"
|
||
}
|
||
},
|
||
"overrides": [
|
||
{}
|
||
],
|
||
"version": "2.0.0",
|
||
"type": "table",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 0,
|
||
"y": 0,
|
||
"i": "98f46bc1-c078-40f2-915c-f0836957bf2f",
|
||
"isResizable": true
|
||
},
|
||
"id": "98f46bc1-c078-40f2-915c-f0836957bf2f"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "apiserver_requested_deprecated_apis{job=\"apiserver\"}",
|
||
"legend": ""
|
||
}
|
||
],
|
||
"name": "Deprecated Kubernetes Resources",
|
||
"links": [],
|
||
"description": "当前版本apiserver使用，未来版本中要移除的资源",
|
||
"custom": {
|
||
"showHeader": true,
|
||
"colorMode": "value",
|
||
"calc": "lastNotNull",
|
||
"displayMode": "labelsOfSeriesToRows",
|
||
"columns": [
|
||
"group",
|
||
"version",
|
||
"resource",
|
||
"removed_release"
|
||
],
|
||
"sortOrder": "ascend"
|
||
},
|
||
"options": {
|
||
"standardOptions": {}
|
||
},
|
||
"overrides": [
|
||
{}
|
||
],
|
||
"version": "2.0.0",
|
||
"type": "table",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 12,
|
||
"y": 0,
|
||
"i": "73beb13a-bd10-4a68-bb9e-5b9ab63da154",
|
||
"isResizable": true
|
||
},
|
||
"id": "73beb13a-bd10-4a68-bb9e-5b9ab63da154"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "sum by (instance,code) (rate(apiserver_request_total{job=\"apiserver\"}[5m]))",
|
||
"legend": "{{ instance }} {{ code }}"
|
||
}
|
||
],
|
||
"name": "API Server - HTTP Requests by code",
|
||
"links": [],
|
||
"description": "按照返回码分类统计apiserver请求数",
|
||
"options": {
|
||
"tooltip": {
|
||
"mode": "all",
|
||
"sort": "none"
|
||
},
|
||
"legend": {
|
||
"displayMode": "list",
|
||
"placement": "bottom"
|
||
},
|
||
"standardOptions": {
|
||
"util": "none"
|
||
},
|
||
"thresholds": {
|
||
"steps": []
|
||
}
|
||
},
|
||
"custom": {
|
||
"drawStyle": "lines",
|
||
"lineInterpolation": "smooth",
|
||
"lineWidth": 1,
|
||
"fillOpacity": 0.25,
|
||
"gradientMode": "none",
|
||
"stack": "off"
|
||
},
|
||
"version": "2.0.0",
|
||
"type": "timeseries",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 0,
|
||
"y": 8,
|
||
"i": "1cfa42b1-9dcf-471c-90ff-8ffe656d4b11",
|
||
"isResizable": true
|
||
},
|
||
"id": "1cfa42b1-9dcf-471c-90ff-8ffe656d4b11"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "sum by (instance,verb) (rate(apiserver_request_total{job=\"apiserver\"}[5m]))",
|
||
"legend": "{{ instance }} {{ verb }}"
|
||
}
|
||
],
|
||
"name": "API Server - HTTP Requests by verb",
|
||
"links": [],
|
||
"description": "按照请求动作分类统计apiserver的请求数",
|
||
"options": {
|
||
"tooltip": {
|
||
"mode": "all",
|
||
"sort": "desc"
|
||
},
|
||
"legend": {
|
||
"displayMode": "list",
|
||
"placement": "bottom"
|
||
},
|
||
"standardOptions": {
|
||
"util": "none"
|
||
},
|
||
"thresholds": {
|
||
"steps": []
|
||
}
|
||
},
|
||
"custom": {
|
||
"drawStyle": "lines",
|
||
"lineInterpolation": "smooth",
|
||
"lineWidth": 1,
|
||
"fillOpacity": 0.25,
|
||
"gradientMode": "none",
|
||
"stack": "off"
|
||
},
|
||
"version": "2.0.0",
|
||
"type": "timeseries",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 12,
|
||
"y": 8,
|
||
"i": "94def0cb-0b86-42f7-a4b2-dde714bbb918",
|
||
"isResizable": true
|
||
},
|
||
"id": "94def0cb-0b86-42f7-a4b2-dde714bbb918"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "apiserver_current_inflight_requests{job=\"apiserver\"}",
|
||
"legend": "{{ instance }} {{ request_kind }}"
|
||
}
|
||
],
|
||
"name": "API Server - Current Inflight Requests by kind",
|
||
"links": [],
|
||
"description": "当前并发请求apiserver的数量",
|
||
"options": {
|
||
"tooltip": {
|
||
"mode": "all",
|
||
"sort": "desc"
|
||
},
|
||
"legend": {
|
||
"displayMode": "list",
|
||
"placement": "bottom"
|
||
},
|
||
"standardOptions": {
|
||
"util": "none"
|
||
},
|
||
"thresholds": {
|
||
"steps": []
|
||
}
|
||
},
|
||
"custom": {
|
||
"drawStyle": "lines",
|
||
"lineInterpolation": "smooth",
|
||
"lineWidth": 1,
|
||
"fillOpacity": 0.25,
|
||
"gradientMode": "none",
|
||
"stack": "off"
|
||
},
|
||
"version": "2.0.0",
|
||
"type": "timeseries",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 0,
|
||
"y": 16,
|
||
"i": "ce5a15ad-11c6-44a2-a071-be57009162e1",
|
||
"isResizable": true
|
||
},
|
||
"id": "fb6266a3-3da0-4310-bfe8-c64a53db5db3"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "histogram_quantile(0.9, sum(rate(apiserver_request_duration_seconds_bucket{job=\"apiserver\"}[5m])) by (instance,verb,le))*1000",
|
||
"legend": "{{ instance }} {{ verb }}"
|
||
}
|
||
],
|
||
"name": "API Server - HTTP Requets Latency by verb",
|
||
"links": [],
|
||
"description": "apiserver的响应延迟，按请求动作分类统计",
|
||
"options": {
|
||
"tooltip": {
|
||
"mode": "all",
|
||
"sort": "none"
|
||
},
|
||
"legend": {
|
||
"displayMode": "list",
|
||
"placement": "bottom"
|
||
},
|
||
"standardOptions": {
|
||
"util": "milliseconds"
|
||
},
|
||
"thresholds": {
|
||
"steps": []
|
||
}
|
||
},
|
||
"custom": {
|
||
"drawStyle": "lines",
|
||
"lineInterpolation": "smooth",
|
||
"lineWidth": 1,
|
||
"fillOpacity": 0.25,
|
||
"gradientMode": "none",
|
||
"stack": "off"
|
||
},
|
||
"version": "2.0.0",
|
||
"type": "timeseries",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 12,
|
||
"y": 16,
|
||
"i": "045dca2d-d69b-47a7-b25e-656adb357e11",
|
||
"isResizable": true
|
||
},
|
||
"id": "045dca2d-d69b-47a7-b25e-656adb357e11"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "histogram_quantile(0.9, sum(rate(apiserver_request_duration_seconds_bucket {job=\"apiserver\",verb!=\"WATCH\"}[5m])) by (instance,le))*1000",
|
||
"legend": "{{ instance }}"
|
||
}
|
||
],
|
||
"name": "API Server - HTTP Requets Latency by instance",
|
||
"links": [],
|
||
"description": "apiserver的响应延迟(非watch请求）",
|
||
"options": {
|
||
"tooltip": {
|
||
"mode": "all",
|
||
"sort": "desc"
|
||
},
|
||
"legend": {
|
||
"displayMode": "list",
|
||
"placement": "bottom"
|
||
},
|
||
"standardOptions": {
|
||
"util": "milliseconds"
|
||
},
|
||
"thresholds": {
|
||
"steps": []
|
||
}
|
||
},
|
||
"custom": {
|
||
"drawStyle": "lines",
|
||
"lineInterpolation": "smooth",
|
||
"lineWidth": 1,
|
||
"fillOpacity": 0.25,
|
||
"gradientMode": "none",
|
||
"stack": "off"
|
||
},
|
||
"version": "2.0.0",
|
||
"type": "timeseries",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 0,
|
||
"y": 24,
|
||
"i": "1e775704-9ee4-45ce-9d24-b49af89fb5c7",
|
||
"isResizable": true
|
||
},
|
||
"id": "1e775704-9ee4-45ce-9d24-b49af89fb5c7"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "sum by(instance,verb) (rate(apiserver_request_total{code=~\"5..\",job=\"apiserver\"}[5m]))\n / sum by(instance,verb) (rate(apiserver_request_total{job=\"apiserver\"}[5m]))",
|
||
"legend": "{{ instance }} {{ verb }}"
|
||
}
|
||
],
|
||
"name": "API Server - Errors by verb",
|
||
"links": [],
|
||
"description": "apiserver的5xx错误率，按请求动作分类统计",
|
||
"options": {
|
||
"tooltip": {
|
||
"mode": "all",
|
||
"sort": "desc"
|
||
},
|
||
"legend": {
|
||
"displayMode": "list",
|
||
"placement": "bottom"
|
||
},
|
||
"standardOptions": {
|
||
"util": "none"
|
||
},
|
||
"thresholds": {
|
||
"steps": []
|
||
}
|
||
},
|
||
"custom": {
|
||
"drawStyle": "lines",
|
||
"lineInterpolation": "smooth",
|
||
"lineWidth": 1,
|
||
"fillOpacity": 0.25,
|
||
"gradientMode": "none",
|
||
"stack": "off"
|
||
},
|
||
"version": "2.0.0",
|
||
"type": "timeseries",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 12,
|
||
"y": 24,
|
||
"i": "1ca62e0b-72df-47d1-93ba-048ed49e9cb5",
|
||
"isResizable": true
|
||
},
|
||
"id": "1ca62e0b-72df-47d1-93ba-048ed49e9cb5"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "sum by(instance) (rate(apiserver_request_total{code=~\"5..\", job=\"apiserver\"}[5m]))\n / sum by(instance) (rate(apiserver_request_total{job=\"apiserver\"}[5m]))",
|
||
"legend": "{{ instance }}"
|
||
}
|
||
],
|
||
"name": "API Server - Errors by Instance",
|
||
"links": [],
|
||
"description": "apiserver的5xx 错误率(5xx请求数/总请求数)",
|
||
"options": {
|
||
"tooltip": {
|
||
"mode": "all",
|
||
"sort": "desc"
|
||
},
|
||
"legend": {
|
||
"displayMode": "list",
|
||
"placement": "bottom"
|
||
},
|
||
"standardOptions": {
|
||
"util": "none"
|
||
},
|
||
"thresholds": {
|
||
"steps": []
|
||
}
|
||
},
|
||
"custom": {
|
||
"drawStyle": "lines",
|
||
"lineInterpolation": "smooth",
|
||
"lineWidth": 1,
|
||
"fillOpacity": 0.25,
|
||
"gradientMode": "none",
|
||
"stack": "off"
|
||
},
|
||
"version": "2.0.0",
|
||
"type": "timeseries",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 0,
|
||
"y": 32,
|
||
"i": "92a209a1-7d30-4627-9ae1-55ded5095ed7",
|
||
"isResizable": true
|
||
},
|
||
"id": "92a209a1-7d30-4627-9ae1-55ded5095ed7"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "sum(rate(workqueue_depth{job=\"apiserver\"}[5m])) by (instance,name)",
|
||
"legend": "{{ instance }} {{ name }}"
|
||
}
|
||
],
|
||
"name": "API Server - Work Queue by instance",
|
||
"links": [],
|
||
"description": "apiserver工作队列深度，越接近0越好",
|
||
"options": {
|
||
"tooltip": {
|
||
"mode": "all",
|
||
"sort": "desc"
|
||
},
|
||
"legend": {
|
||
"displayMode": "list",
|
||
"placement": "bottom"
|
||
},
|
||
"standardOptions": {
|
||
"util": "none"
|
||
},
|
||
"thresholds": {
|
||
"steps": []
|
||
}
|
||
},
|
||
"custom": {
|
||
"drawStyle": "lines",
|
||
"lineInterpolation": "smooth",
|
||
"lineWidth": 1,
|
||
"fillOpacity": 0.25,
|
||
"gradientMode": "none",
|
||
"stack": "off"
|
||
},
|
||
"version": "2.0.0",
|
||
"type": "timeseries",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 12,
|
||
"y": 32,
|
||
"i": "83f22cf4-9c65-4ad3-900b-fa6fc914dd88",
|
||
"isResizable": true
|
||
},
|
||
"id": "83f22cf4-9c65-4ad3-900b-fa6fc914dd88"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "sum(rate(apiserver_request_total{job=\"apiserver\"}[5m])) by (instance)",
|
||
"legend": "{{ instance }}"
|
||
}
|
||
],
|
||
"name": "API Server - HTTP Requests by instance",
|
||
"links": [],
|
||
"description": "5分钟内apiserver的请求数统计",
|
||
"options": {
|
||
"tooltip": {
|
||
"mode": "all",
|
||
"sort": "desc"
|
||
},
|
||
"legend": {
|
||
"displayMode": "list",
|
||
"placement": "bottom"
|
||
},
|
||
"standardOptions": {
|
||
"util": "none"
|
||
},
|
||
"thresholds": {
|
||
"steps": []
|
||
}
|
||
},
|
||
"custom": {
|
||
"drawStyle": "lines",
|
||
"lineInterpolation": "smooth",
|
||
"lineWidth": 1,
|
||
"fillOpacity": 0.25,
|
||
"gradientMode": "none",
|
||
"stack": "normal"
|
||
},
|
||
"version": "2.0.0",
|
||
"type": "timeseries",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 0,
|
||
"y": 40,
|
||
"i": "3e9f9df7-d9fb-4791-b3b2-2c52678f060f",
|
||
"isResizable": true
|
||
},
|
||
"id": "3e9f9df7-d9fb-4791-b3b2-2c52678f060f"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "rate(process_cpu_seconds_total{job=\"apiserver\"}[5m])",
|
||
"legend": "{{ instance }}"
|
||
}
|
||
],
|
||
"name": "API Server - CPU Usage by instance",
|
||
"links": [],
|
||
"description": "apiserver的cpu使用率",
|
||
"options": {
|
||
"tooltip": {
|
||
"mode": "all",
|
||
"sort": "desc"
|
||
},
|
||
"legend": {
|
||
"displayMode": "list",
|
||
"placement": "bottom"
|
||
},
|
||
"standardOptions": {
|
||
"util": "percent",
|
||
"decimals": 2
|
||
},
|
||
"thresholds": {
|
||
"steps": []
|
||
}
|
||
},
|
||
"custom": {
|
||
"drawStyle": "lines",
|
||
"lineInterpolation": "smooth",
|
||
"lineWidth": 1,
|
||
"fillOpacity": 0.25,
|
||
"gradientMode": "none",
|
||
"stack": "off"
|
||
},
|
||
"version": "2.0.0",
|
||
"type": "timeseries",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 12,
|
||
"y": 40,
|
||
"i": "3d5c1ae5-e640-4986-9202-78258169bffb",
|
||
"isResizable": true
|
||
},
|
||
"id": "3d5c1ae5-e640-4986-9202-78258169bffb"
|
||
},
|
||
{
|
||
"targets": [
|
||
{
|
||
"refId": "A",
|
||
"expr": "process_resident_memory_bytes{job=\"apiserver\"}",
|
||
"legend": "{{ instance }}"
|
||
}
|
||
],
|
||
"name": "API Server - Memory Usage by instance",
|
||
"links": [],
|
||
"description": "apiserver的内存使用量",
|
||
"options": {
|
||
"tooltip": {
|
||
"mode": "all",
|
||
"sort": "desc"
|
||
},
|
||
"legend": {
|
||
"displayMode": "list",
|
||
"placement": "bottom"
|
||
},
|
||
"standardOptions": {
|
||
"util": "bytesIEC"
|
||
},
|
||
"thresholds": {
|
||
"steps": []
|
||
}
|
||
},
|
||
"custom": {
|
||
"drawStyle": "lines",
|
||
"lineInterpolation": "smooth",
|
||
"lineWidth": 1,
|
||
"fillOpacity": 0.25,
|
||
"gradientMode": "none",
|
||
"stack": "off"
|
||
},
|
||
"version": "2.0.0",
|
||
"type": "timeseries",
|
||
"layout": {
|
||
"h": 8,
|
||
"w": 12,
|
||
"x": 0,
|
||
"y": 48,
|
||
"i": "1550a2d5-c808-4174-865a-a41b2c16b486",
|
||
"isResizable": true
|
||
},
|
||
"id": "1550a2d5-c808-4174-865a-a41b2c16b486"
|
||
}
|
||
]
|
||
}
|
||
}
|