ccfos
/
nightingale
Not watched
1
0
Fork 0
Code
Releases 20 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
Tree: 56a7860b5a
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '56a7860b5a'
${ noResults }
nightingale/integrations/Process/dashboards/process_by_exporter.json

1106 lines
44 kB
Raw Blame History 

  1. {

  2.     "name": "Linux Process - exporter",

  3.     "tags": "Prometheus Process",

  4.     "ident": "",

  5.     "configs": {

  6.         "panels": [

  7.             {

  8.                 "collapsed": true,

  9.                 "id": "021aedd5-dac7-4431-aeea-50c0c594e784",

  10.                 "layout": {

  11.                     "h": 1,

  12.                     "i": "021aedd5-dac7-4431-aeea-50c0c594e784",

  13.                     "w": 24,

  14.                     "x": 0,

  15.                     "y": 0

  16.                 },

  17.                 "name": "Cpu Usage",

  18.                 "type": "row"

  19.             },

  20.             {

  21.                 "custom": {

  22.                     "drawStyle": "lines",

  23.                     "fillOpacity": 0.5,

  24.                     "lineInterpolation": "smooth",

  25.                     "stack": "off"

  26.                 },

  27.                 "description": "进程占用CPU时间（用户态+内核态），倒排前5",

  28.                 "id": "0f55cad5-ae67-44de-bba5-f6e1bd23066a",

  29.                 "layout": {

  30.                     "h": 7,

  31.                     "i": "0f55cad5-ae67-44de-bba5-f6e1bd23066a",

  32.                     "w": 12,

  33.                     "x": 0,

  34.                     "y": 1

  35.                 },

  36.                 "name": "Top processes by Total CPU cores used",

  37.                 "options": {

  38.                     "legend": {

  39.                         "displayMode": "hidden"

  40.                     },

  41.                     "standardOptions": {

  42.                         "util": "percentUnit"

  43.                     },

  44.                     "thresholds": {},

  45.                     "tooltip": {

  46.                         "mode": "all",

  47.                         "sort": "none"

  48.                     }

  49.                 },

  50.                 "targets": [

  51.                     {

  52.                         "expr": "topk(5,(rate(namedprocess_namegroup_cpu_seconds_total{mode=\"user\",groupname=~\"$processes\",instance=~\"$instance\"}[5m])  + ignoring(mode)  rate(namedprocess_namegroup_cpu_seconds_total{mode=\"system\",groupname=~\"$processes\",instance=~\"$instance\"}[5m])) or  (irate(namedprocess_namegroup_cpu_seconds_total{mode=\"user\",groupname=~\"$processes\",instance=~\"$instance\"}[5m]) + ignoring(mode) irate(namedprocess_namegroup_cpu_seconds_total{mode=\"system\",groupname=~\"$processes\",instance=~\"$instance\"}[5m])))",

  53.                         "legend": "{{groupname}}",

  54.                         "refId": "A"

  55.                     }

  56.                 ],

  57.                 "type": "timeseries",

  58.                 "version": "2.0.0",

  59.                 "datasourceCate": "prometheus",

  60.                 "datasourceValue": "${prom}"

  61.             },

  62.             {

  63.                 "custom": {

  64.                     "drawStyle": "lines",

  65.                     "fillOpacity": 0.5,

  66.                     "lineInterpolation": "smooth",

  67.                     "stack": "off"

  68.                 },

  69.                 "description": "进程占用CPU时间（内核态），倒排前5",

  70.                 "id": "e200eb2a-9cd6-4954-880f-9a354c1af2cc",

  71.                 "layout": {

  72.                     "h": 7,

  73.                     "i": "e200eb2a-9cd6-4954-880f-9a354c1af2cc",

  74.                     "w": 12,

  75.                     "x": 12,

  76.                     "y": 1

  77.                 },

  78.                 "name": "Top processes by System CPU cores used",

  79.                 "options": {

  80.                     "legend": {

  81.                         "displayMode": "hidden"

  82.                     },

  83.                     "standardOptions": {

  84.                         "util": "percentUnit"

  85.                     },

  86.                     "thresholds": {},

  87.                     "tooltip": {

  88.                         "mode": "all",

  89.                         "sort": "none"

  90.                     }

  91.                 },

  92.                 "targets": [

  93.                     {

  94.                         "expr": "topk(5, rate(namedprocess_namegroup_cpu_seconds_total{mode=\"system\",groupname=~\"$processes\",instance=~\"$instance\"}[5m]) or  ( irate(namedprocess_namegroup_cpu_seconds_total{mode=\"system\",groupname=~\"$processes\",instance=~\"$instance\"}[5m])))",

  95.                         "legend": "{{groupname}}",

  96.                         "refId": "A"

  97.                     }

  98.                 ],

  99.                 "type": "timeseries",

  100.                 "version": "2.0.0",

  101.                 "datasourceCate": "prometheus",

  102.                 "datasourceValue": "${prom}"

  103.             },

  104.             {

  105.                 "collapsed": true,

  106.                 "id": "3f0a9068-51ab-4889-82e1-444cd23c56e8",

  107.                 "layout": {

  108.                     "h": 1,

  109.                     "i": "3f0a9068-51ab-4889-82e1-444cd23c56e8",

  110.                     "w": 24,

  111.                     "x": 0,

  112.                     "y": 8

  113.                 },

  114.                 "name": "Memory Usage",

  115.                 "type": "row"

  116.             },

  117.             {

  118.                 "custom": {

  119.                     "drawStyle": "lines",

  120.                     "fillOpacity": 0.5,

  121.                     "lineInterpolation": "smooth",

  122.                     "stack": "off"

  123.                 },

  124.                 "description": "进程常驻内存与交换空间平均占用容量之和，倒排前5",

  125.                 "id": "9cb0cc2f-c122-4f5e-98ed-8bf310bd612a",

  126.                 "layout": {

  127.                     "h": 7,

  128.                     "i": "9cb0cc2f-c122-4f5e-98ed-8bf310bd612a",

  129.                     "w": 12,

  130.                     "x": 0,

  131.                     "y": 9

  132.                 },

  133.                 "name": "Top processes by Used memory",

  134.                 "options": {

  135.                     "legend": {

  136.                         "displayMode": "hidden"

  137.                     },

  138.                     "standardOptions": {

  139.                         "util": "bytesIEC"

  140.                     },

  141.                     "thresholds": {},

  142.                     "tooltip": {

  143.                         "mode": "all",

  144.                         "sort": "none"

  145.                     }

  146.                 },

  147.                 "targets": [

  148.                     {

  149.                         "expr": "topk(5,(  (avg_over_time(namedprocess_namegroup_memory_bytes{groupname=~\"$processes\", memtype=\"swapped\",instance=~\"$instance\"}[5m])+ ignoring (memtype) avg_over_time(namedprocess_namegroup_memory_bytes{groupname=~\"$processes\", memtype=\"resident\",instance=~\"$instance\"}[5m]))  or  (avg_over_time(namedprocess_namegroup_memory_bytes{groupname=~\"$processes\", memtype=\"swapped\",instance=~\"$instance\"}[5m])+ ignoring (memtype) avg_over_time(namedprocess_namegroup_memory_bytes{groupname=~\"$processes\", memtype=\"resident\",instance=~\"$instance\"}[5m]))  ))",

  150.                         "legend": "{{groupname}}",

  151.                         "refId": "A"

  152.                     }

  153.                 ],

  154.                 "type": "timeseries",

  155.                 "version": "2.0.0",

  156.                 "datasourceCate": "prometheus",

  157.                 "datasourceValue": "${prom}"

  158.             },

  159.             {

  160.                 "custom": {

  161.                     "drawStyle": "lines",

  162.                     "fillOpacity": 0.5,

  163.                     "lineInterpolation": "smooth",

  164.                     "stack": "off"

  165.                 },

  166.                 "description": "进程常驻内存平均占用容量之和，倒排前5",

  167.                 "id": "7aceae00-23be-447f-87df-47f74b278190",

  168.                 "layout": {

  169.                     "h": 7,

  170.                     "i": "7aceae00-23be-447f-87df-47f74b278190",

  171.                     "w": 12,

  172.                     "x": 12,

  173.                     "y": 9

  174.                 },

  175.                 "name": "Top processes by Resident Memory",

  176.                 "options": {

  177.                     "legend": {

  178.                         "displayMode": "hidden"

  179.                     },

  180.                     "standardOptions": {

  181.                         "util": "bytesIEC"

  182.                     },

  183.                     "thresholds": {},

  184.                     "tooltip": {

  185.                         "mode": "all",

  186.                         "sort": "none"

  187.                     }

  188.                 },

  189.                 "targets": [

  190.                     {

  191.                         "expr": "topk(5, (avg_over_time(namedprocess_namegroup_memory_bytes{groupname=~\"$processes\", memtype=\"resident\",instance=~\"$instance\"}[5m])  or avg_over_time(namedprocess_namegroup_memory_bytes{groupname=~\"$processes\", memtype=\"resident\",instance=~\"$instance\"}[5m]) ))",

  192.                         "legend": "{{groupname}}",

  193.                         "refId": "A"

  194.                     }

  195.                 ],

  196.                 "type": "timeseries",

  197.                 "version": "2.0.0",

  198.                 "datasourceCate": "prometheus",

  199.                 "datasourceValue": "${prom}"

  200.             },

  201.             {

  202.                 "custom": {

  203.                     "drawStyle": "lines",

  204.                     "fillOpacity": 0.5,

  205.                     "lineInterpolation": "smooth",

  206.                     "stack": "off"

  207.                 },

  208.                 "description": "进程交换内存平均占用容量，倒排前5",

  209.                 "id": "55b255d8-756e-421c-9690-675d7c174e86",

  210.                 "layout": {

  211.                     "h": 7,

  212.                     "i": "55b255d8-756e-421c-9690-675d7c174e86",

  213.                     "w": 12,

  214.                     "x": 0,

  215.                     "y": 11

  216.                 },

  217.                 "name": "Top processes by Swapped Memory",

  218.                 "options": {

  219.                     "legend": {

  220.                         "displayMode": "hidden"

  221.                     },

  222.                     "standardOptions": {

  223.                         "util": "bytesIEC"

  224.                     },

  225.                     "thresholds": {},

  226.                     "tooltip": {

  227.                         "mode": "all",

  228.                         "sort": "none"

  229.                     }

  230.                 },

  231.                 "targets": [

  232.                     {

  233.                         "expr": "topk(5,( avg_over_time(namedprocess_namegroup_memory_bytes{groupname=~\"$processes\", memtype=\"swapped\",instance=~\"$instance\"}[5m]) or avg_over_time(namedprocess_namegroup_memory_bytes{groupname=~\"$processes\", memtype=\"swapped\",instance=~\"$instance\"}[5m]))) ",

  234.                         "legend": "{{groupname}}",

  235.                         "refId": "A"

  236.                     }

  237.                 ],

  238.                 "type": "timeseries",

  239.                 "version": "2.0.0",

  240.                 "datasourceCate": "prometheus",

  241.                 "datasourceValue": "${prom}"

  242.             },

  243.             {

  244.                 "custom": {

  245.                     "drawStyle": "lines",

  246.                     "fillOpacity": 0.5,

  247.                     "lineInterpolation": "smooth",

  248.                     "stack": "off"

  249.                 },

  250.                 "description": "进程虚拟内存平均占用容量，倒排前5",

  251.                 "id": "3c900819-5955-4bef-b419-ea8c47107b0d",

  252.                 "layout": {

  253.                     "h": 7,

  254.                     "i": "3c900819-5955-4bef-b419-ea8c47107b0d",

  255.                     "w": 12,

  256.                     "x": 12,

  257.                     "y": 11

  258.                 },

  259.                 "name": "Top processes by Virtual Memory",

  260.                 "options": {

  261.                     "legend": {

  262.                         "displayMode": "hidden"

  263.                     },

  264.                     "standardOptions": {

  265.                         "util": "bytesIEC"

  266.                     },

  267.                     "thresholds": {},

  268.                     "tooltip": {

  269.                         "mode": "all",

  270.                         "sort": "none"

  271.                     }

  272.                 },

  273.                 "targets": [

  274.                     {

  275.                         "expr": "topk(5,( avg_over_time(namedprocess_namegroup_memory_bytes{groupname=~\"$processes\", memtype=\"virtual\",instance=~\"$instance\"}[5m]) or avg_over_time(namedprocess_namegroup_memory_bytes{groupname=~\"$processes\", memtype=\"virtual\",instance=~\"$instance\"}[5m])))",

  276.                         "legend": "{{groupname}}",

  277.                         "refId": "A"

  278.                     }

  279.                 ],

  280.                 "type": "timeseries",

  281.                 "version": "2.0.0",

  282.                 "datasourceCate": "prometheus",

  283.                 "datasourceValue": "${prom}"

  284.             },

  285.             {

  286.                 "collapsed": true,

  287.                 "id": "4d72a763-48f3-480b-9701-bc297be1bbfb",

  288.                 "layout": {

  289.                     "h": 1,

  290.                     "i": "4d72a763-48f3-480b-9701-bc297be1bbfb",

  291.                     "w": 24,

  292.                     "x": 0,

  293.                     "y": 18

  294.                 },

  295.                 "name": "Disk IO Usage",

  296.                 "type": "row"

  297.             },

  298.             {

  299.                 "custom": {

  300.                     "drawStyle": "lines",

  301.                     "fillOpacity": 0.5,

  302.                     "lineInterpolation": "smooth",

  303.                     "stack": "off"

  304.                 },

  305.                 "description": "进程写数据量，倒排前5",

  306.                 "id": "4cb30905-65f6-42aa-895e-ef3679905ef8",

  307.                 "layout": {

  308.                     "h": 7,

  309.                     "i": "4cb30905-65f6-42aa-895e-ef3679905ef8",

  310.                     "w": 12,

  311.                     "x": 0,

  312.                     "y": 19

  313.                 },

  314.                 "name": "Top processes by Bytes Written",

  315.                 "options": {

  316.                     "legend": {

  317.                         "displayMode": "hidden"

  318.                     },

  319.                     "standardOptions": {

  320.                         "util": "bytesSI"

  321.                     },

  322.                     "thresholds": {},

  323.                     "tooltip": {

  324.                         "mode": "all",

  325.                         "sort": "none"

  326.                     }

  327.                 },

  328.                 "targets": [

  329.                     {

  330.                         "expr": "topk(5,(rate(namedprocess_namegroup_write_bytes_total{groupname=~\"$processes\",instance=~\"$instance\"}[5m]) or irate(namedprocess_namegroup_write_bytes_total{groupname=~\"$processes\",instance=~\"$instance\"}[5m])))",

  331.                         "legend": "{{groupname}}",

  332.                         "refId": "A"

  333.                     }

  334.                 ],

  335.                 "type": "timeseries",

  336.                 "version": "2.0.0",

  337.                 "datasourceCate": "prometheus",

  338.                 "datasourceValue": "${prom}"

  339.             },

  340.             {

  341.                 "custom": {

  342.                     "drawStyle": "lines",

  343.                     "fillOpacity": 0.5,

  344.                     "lineInterpolation": "smooth",

  345.                     "stack": "off"

  346.                 },

  347.                 "description": "进程读数据量，倒排前5",

  348.                 "id": "d74f4d5f-db71-4781-b539-0df17dfcd42f",

  349.                 "layout": {

  350.                     "h": 7,

  351.                     "i": "d74f4d5f-db71-4781-b539-0df17dfcd42f",

  352.                     "w": 12,

  353.                     "x": 12,

  354.                     "y": 19

  355.                 },

  356.                 "name": "Top processes by Bytes Read",

  357.                 "options": {

  358.                     "legend": {

  359.                         "displayMode": "hidden"

  360.                     },

  361.                     "standardOptions": {

  362.                         "util": "bytesSI"

  363.                     },

  364.                     "thresholds": {},

  365.                     "tooltip": {

  366.                         "mode": "all",

  367.                         "sort": "none"

  368.                     }

  369.                 },

  370.                 "targets": [

  371.                     {

  372.                         "expr": "topk(5,(rate(namedprocess_namegroup_read_bytes_total{groupname=~\"$processes\",instance=~\"$instance\"}[5m]) or irate(namedprocess_namegroup_read_bytes_total{groupname=~\"$processes\",instance=~\"$instance\"}[5m])))",

  373.                         "legend": "{{groupname}}",

  374.                         "refId": "A"

  375.                     }

  376.                 ],

  377.                 "type": "timeseries",

  378.                 "version": "2.0.0",

  379.                 "datasourceCate": "prometheus",

  380.                 "datasourceValue": "${prom}"

  381.             },

  382.             {

  383.                 "collapsed": true,

  384.                 "id": "e460097e-c3c8-45da-8202-dee1de996ddf",

  385.                 "layout": {

  386.                     "h": 1,

  387.                     "i": "e460097e-c3c8-45da-8202-dee1de996ddf",

  388.                     "w": 24,

  389.                     "x": 0,

  390.                     "y": 26

  391.                 },

  392.                 "name": "Process and Thread Counts",

  393.                 "type": "row"

  394.             },

  395.             {

  396.                 "custom": {

  397.                     "drawStyle": "lines",

  398.                     "fillOpacity": 0.5,

  399.                     "lineInterpolation": "smooth",

  400.                     "stack": "off"

  401.                 },

  402.                 "description": "同名进程数，倒排前5",

  403.                 "id": "1b44ce6f-ca65-49f5-892e-05444c9aa580",

  404.                 "layout": {

  405.                     "h": 7,

  406.                     "i": "1b44ce6f-ca65-49f5-892e-05444c9aa580",

  407.                     "w": 12,

  408.                     "x": 0,

  409.                     "y": 27

  410.                 },

  411.                 "name": "Top processes by number of processes instances",

  412.                 "options": {

  413.                     "legend": {

  414.                         "displayMode": "hidden"

  415.                     },

  416.                     "standardOptions": {},

  417.                     "thresholds": {},

  418.                     "tooltip": {

  419.                         "mode": "all",

  420.                         "sort": "none"

  421.                     }

  422.                 },

  423.                 "targets": [

  424.                     {

  425.                         "expr": "topk(5,(max_over_time(namedprocess_namegroup_num_procs{groupname=~\"$processes\",instance=~\"$instance\"}[5m])  or max_over_time(namedprocess_namegroup_num_procs{groupname=~\"$processes\",instance=~\"$instance\"}[5m])))",

  426.                         "legend": "{{groupname}}",

  427.                         "refId": "A"

  428.                     }

  429.                 ],

  430.                 "type": "timeseries",

  431.                 "version": "2.0.0",

  432.                 "datasourceCate": "prometheus",

  433.                 "datasourceValue": "${prom}"

  434.             },

  435.             {

  436.                 "custom": {

  437.                     "drawStyle": "lines",

  438.                     "fillOpacity": 0.5,

  439.                     "lineInterpolation": "smooth",

  440.                     "stack": "off"

  441.                 },

  442.                 "description": "进程内线程数，倒排前5",

  443.                 "id": "24975aae-a573-4998-bdab-ffcb19cee2ae",

  444.                 "layout": {

  445.                     "h": 7,

  446.                     "i": "24975aae-a573-4998-bdab-ffcb19cee2ae",

  447.                     "w": 12,

  448.                     "x": 12,

  449.                     "y": 27

  450.                 },

  451.                 "name": "Top processes by number of threads",

  452.                 "options": {

  453.                     "legend": {

  454.                         "displayMode": "hidden"

  455.                     },

  456.                     "standardOptions": {},

  457.                     "thresholds": {},

  458.                     "tooltip": {

  459.                         "mode": "all",

  460.                         "sort": "none"

  461.                     }

  462.                 },

  463.                 "targets": [

  464.                     {

  465.                         "expr": "topk(5,(max_over_time(namedprocess_namegroup_num_threads{groupname=~\"$processes\",instance=~\"$instance\"}[5m]) or max_over_time(namedprocess_namegroup_num_threads{groupname=~\"$processes\",instance=~\"$instance\"}[5m])))",

  466.                         "legend": "{{groupname}}",

  467.                         "refId": "A"

  468.                     }

  469.                 ],

  470.                 "type": "timeseries",

  471.                 "version": "2.0.0",

  472.                 "datasourceCate": "prometheus",

  473.                 "datasourceValue": "${prom}"

  474.             },

  475.             {

  476.                 "collapsed": true,

  477.                 "id": "3df9fc7b-bf9c-45bc-baa1-72ef83bc6b34",

  478.                 "layout": {

  479.                     "h": 1,

  480.                     "i": "3df9fc7b-bf9c-45bc-baa1-72ef83bc6b34",

  481.                     "w": 24,

  482.                     "x": 0,

  483.                     "y": 34

  484.                 },

  485.                 "name": "Context Switches",

  486.                 "type": "row"

  487.             },

  488.             {

  489.                 "custom": {

  490.                     "drawStyle": "lines",

  491.                     "fillOpacity": 0.5,

  492.                     "lineInterpolation": "smooth",

  493.                     "stack": "off"

  494.                 },

  495.                 "description": "进程自愿中断（如I/O完成）次数，倒排前5",

  496.                 "id": "3775f2e7-fa98-4458-9659-0d1221b3cc03",

  497.                 "layout": {

  498.                     "h": 7,

  499.                     "i": "3775f2e7-fa98-4458-9659-0d1221b3cc03",

  500.                     "w": 12,

  501.                     "x": 0,

  502.                     "y": 35

  503.                 },

  504.                 "name": "Top Processes by Voluntary Context Switches",

  505.                 "options": {

  506.                     "legend": {

  507.                         "displayMode": "hidden"

  508.                     },

  509.                     "standardOptions": {},

  510.                     "thresholds": {},

  511.                     "tooltip": {

  512.                         "mode": "all",

  513.                         "sort": "none"

  514.                     }

  515.                 },

  516.                 "targets": [

  517.                     {

  518.                         "expr": "topk(5,( rate(namedprocess_namegroup_context_switches_total{groupname=~\"$processes\",instance=~\"$instance\",ctxswitchtype=\"voluntary\"}[5m]) or irate(namedprocess_namegroup_context_switches_total{groupname=~\"$processes\",instance=~\"$instance\",ctxswitchtype=\"voluntary\"}[5m])))",

  519.                         "legend": "{{groupname}}",

  520.                         "refId": "A"

  521.                     }

  522.                 ],

  523.                 "type": "timeseries",

  524.                 "version": "2.0.0",

  525.                 "datasourceCate": "prometheus",

  526.                 "datasourceValue": "${prom}"

  527.             },

  528.             {

  529.                 "custom": {

  530.                     "drawStyle": "lines",

  531.                     "fillOpacity": 0.5,

  532.                     "lineInterpolation": "smooth",

  533.                     "stack": "off"

  534.                 },

  535.                 "description": "进程被迫中断（如CPU时间耗尽）次数，倒排前5",

  536.                 "id": "a0e912d2-5b0d-42d0-bcf6-27bd54b5f07d",

  537.                 "layout": {

  538.                     "h": 7,

  539.                     "i": "a0e912d2-5b0d-42d0-bcf6-27bd54b5f07d",

  540.                     "w": 12,

  541.                     "x": 12,

  542.                     "y": 35

  543.                 },

  544.                 "name": "Top Processes by  Non-Voluntary Context Switches",

  545.                 "options": {

  546.                     "legend": {

  547.                         "displayMode": "hidden"

  548.                     },

  549.                     "standardOptions": {},

  550.                     "thresholds": {},

  551.                     "tooltip": {

  552.                         "mode": "all",

  553.                         "sort": "none"

  554.                     }

  555.                 },

  556.                 "targets": [

  557.                     {

  558.                         "expr": "topk(5,( rate(namedprocess_namegroup_context_switches_total{groupname=~\"$processes\",instance=~\"$instance\",ctxswitchtype=\"nonvoluntary\"}[5m]) or irate(namedprocess_namegroup_context_switches_total{groupname=~\"$processes\",instance=~\"$instance\",ctxswitchtype=\"nonvoluntary\"}[5m])))",

  559.                         "legend": "{{groupname}}",

  560.                         "refId": "A"

  561.                     }

  562.                 ],

  563.                 "type": "timeseries",

  564.                 "version": "2.0.0",

  565.                 "datasourceCate": "prometheus",

  566.                 "datasourceValue": "${prom}"

  567.             },

  568.             {

  569.                 "collapsed": true,

  570.                 "id": "79aedbba-da75-4393-b322-1565655017c0",

  571.                 "layout": {

  572.                     "h": 1,

  573.                     "i": "79aedbba-da75-4393-b322-1565655017c0",

  574.                     "w": 24,

  575.                     "x": 0,

  576.                     "y": 42

  577.                 },

  578.                 "name": "File Descriptors",

  579.                 "type": "row"

  580.             },

  581.             {

  582.                 "custom": {

  583.                     "drawStyle": "lines",

  584.                     "fillOpacity": 0.5,

  585.                     "lineInterpolation": "smooth",

  586.                     "stack": "off"

  587.                 },

  588.                 "description": "打开文件数，倒排前5",

  589.                 "id": "73dc1195-8ffd-4c6e-b963-d10cde9ae7ba",

  590.                 "layout": {

  591.                     "h": 7,

  592.                     "i": "73dc1195-8ffd-4c6e-b963-d10cde9ae7ba",

  593.                     "w": 12,

  594.                     "x": 0,

  595.                     "y": 43

  596.                 },

  597.                 "name": "Top processes by Open File Descriptors",

  598.                 "options": {

  599.                     "legend": {

  600.                         "displayMode": "hidden"

  601.                     },

  602.                     "standardOptions": {},

  603.                     "thresholds": {},

  604.                     "tooltip": {

  605.                         "mode": "all",

  606.                         "sort": "none"

  607.                     }

  608.                 },

  609.                 "targets": [

  610.                     {

  611.                         "expr": "topk(5,(max_over_time(namedprocess_namegroup_open_filedesc{groupname=~\"$processes\",instance=~\"$instance\"}[5m]) or max_over_time(namedprocess_namegroup_open_filedesc{groupname=~\"$processes\",instance=~\"$instance\"}[5m])))",

  612.                         "legend": "{{groupname}}",

  613.                         "refId": "A"

  614.                     }

  615.                 ],

  616.                 "type": "timeseries",

  617.                 "version": "2.0.0",

  618.                 "datasourceCate": "prometheus",

  619.                 "datasourceValue": "${prom}"

  620.             },

  621.             {

  622.                 "custom": {

  623.                     "drawStyle": "lines",

  624.                     "fillOpacity": 0.5,

  625.                     "lineInterpolation": "smooth",

  626.                     "stack": "off"

  627.                 },

  628.                 "description": "已打开文件数与允许打开文件数占比，倒排前5",

  629.                 "id": "5ffa8032-5b55-4841-a12e-81c1fc8443dd",

  630.                 "layout": {

  631.                     "h": 7,

  632.                     "i": "5ffa8032-5b55-4841-a12e-81c1fc8443dd",

  633.                     "w": 12,

  634.                     "x": 12,

  635.                     "y": 43

  636.                 },

  637.                 "name": "Top processes by File Descriptor Usage Percent",

  638.                 "options": {

  639.                     "legend": {

  640.                         "displayMode": "hidden"

  641.                     },

  642.                     "standardOptions": {

  643.                         "util": "percent"

  644.                     },

  645.                     "thresholds": {},

  646.                     "tooltip": {

  647.                         "mode": "all",

  648.                         "sort": "none"

  649.                     }

  650.                 },

  651.                 "targets": [

  652.                     {

  653.                         "expr": "topk(5,( max_over_time(namedprocess_namegroup_worst_fd_ratio{groupname=~\"$processes\",instance=~\"$instance\"}[5m]) or max_over_time(namedprocess_namegroup_worst_fd_ratio{groupname=~\"$processes\",instance=~\"$instance\"}[5m]) ))*100",

  654.                         "legend": "{{groupname}}",

  655.                         "refId": "A"

  656.                     }

  657.                 ],

  658.                 "type": "timeseries",

  659.                 "version": "2.0.0",

  660.                 "datasourceCate": "prometheus",

  661.                 "datasourceValue": "${prom}"

  662.             },

  663.             {

  664.                 "collapsed": true,

  665.                 "id": "fe967a6c-63bc-48d6-9632-07da1675581b",

  666.                 "layout": {

  667.                     "h": 1,

  668.                     "i": "fe967a6c-63bc-48d6-9632-07da1675581b",

  669.                     "w": 24,

  670.                     "x": 0,

  671.                     "y": 50

  672.                 },

  673.                 "name": "Page Faults",

  674.                 "type": "row"

  675.             },

  676.             {

  677.                 "custom": {

  678.                     "drawStyle": "lines",

  679.                     "fillOpacity": 0.5,

  680.                     "lineInterpolation": "smooth",

  681.                     "stack": "off"

  682.                 },

  683.                 "description": "主要页缺失次数，倒排前5",

  684.                 "id": "8944f17d-51e6-435d-808a-8301b8539f08",

  685.                 "layout": {

  686.                     "h": 7,

  687.                     "i": "8944f17d-51e6-435d-808a-8301b8539f08",

  688.                     "w": 12,

  689.                     "x": 0,

  690.                     "y": 51

  691.                 },

  692.                 "name": "Top processes by Major Page Faults",

  693.                 "options": {

  694.                     "legend": {

  695.                         "displayMode": "hidden"

  696.                     },

  697.                     "standardOptions": {},

  698.                     "thresholds": {},

  699.                     "tooltip": {

  700.                         "mode": "all",

  701.                         "sort": "none"

  702.                     }

  703.                 },

  704.                 "targets": [

  705.                     {

  706.                         "expr": "topk(5,( rate(namedprocess_namegroup_major_page_faults_total{groupname=~\"$processes\",instance=~\"$instance\"}[5m]) or irate(namedprocess_namegroup_major_page_faults_total{groupname=~\"$processes\",instance=~\"$instance\"}[5m])))",

  707.                         "legend": "{{groupname}}",

  708.                         "refId": "A"

  709.                     }

  710.                 ],

  711.                 "type": "timeseries",

  712.                 "version": "2.0.0",

  713.                 "datasourceCate": "prometheus",

  714.                 "datasourceValue": "${prom}"

  715.             },

  716.             {

  717.                 "custom": {

  718.                     "drawStyle": "lines",

  719.                     "fillOpacity": 0.5,

  720.                     "lineInterpolation": "smooth",

  721.                     "stack": "off"

  722.                 },

  723.                 "description": "次要页缺失次数，倒排前5",

  724.                 "id": "b9bed634-d25a-4542-8fd9-4c63e9db766c",

  725.                 "layout": {

  726.                     "h": 7,

  727.                     "i": "b9bed634-d25a-4542-8fd9-4c63e9db766c",

  728.                     "w": 12,

  729.                     "x": 12,

  730.                     "y": 51

  731.                 },

  732.                 "name": "Top processes by Minor Page Faults",

  733.                 "options": {

  734.                     "legend": {

  735.                         "displayMode": "hidden"

  736.                     },

  737.                     "standardOptions": {},

  738.                     "thresholds": {},

  739.                     "tooltip": {

  740.                         "mode": "all",

  741.                         "sort": "none"

  742.                     }

  743.                 },

  744.                 "targets": [

  745.                     {

  746.                         "expr": "topk(5,( rate(namedprocess_namegroup_minor_page_faults_total{groupname=~\"$processes\",instance=~\"$instance\"}[5m]) or irate(namedprocess_namegroup_minor_page_faults_total{groupname=~\"$processes\",instance=~\"$instance\"}[5m])))",

  747.                         "legend": "{{groupname}}",

  748.                         "refId": "A"

  749.                     }

  750.                 ],

  751.                 "type": "timeseries",

  752.                 "version": "2.0.0",

  753.                 "datasourceCate": "prometheus",

  754.                 "datasourceValue": "${prom}"

  755.             },

  756.             {

  757.                 "collapsed": true,

  758.                 "id": "76910d2f-d2c7-48c1-ad05-9651110f0ff1",

  759.                 "layout": {

  760.                     "h": 1,

  761.                     "i": "76910d2f-d2c7-48c1-ad05-9651110f0ff1",

  762.                     "w": 24,

  763.                     "x": 0,

  764.                     "y": 58

  765.                 },

  766.                 "name": "Statuses",

  767.                 "type": "row"

  768.             },

  769.             {

  770.                 "custom": {

  771.                     "drawStyle": "lines",

  772.                     "fillOpacity": 0.5,

  773.                     "lineInterpolation": "smooth",

  774.                     "stack": "off"

  775.                 },

  776.                 "description": "运行态同名进程数量，倒排前5",

  777.                 "id": "7c0a07a2-f936-41c1-a03e-328d8bc1feea",

  778.                 "layout": {

  779.                     "h": 7,

  780.                     "i": "7c0a07a2-f936-41c1-a03e-328d8bc1feea",

  781.                     "w": 12,

  782.                     "x": 0,

  783.                     "y": 59

  784.                 },

  785.                 "name": "Top running processes",

  786.                 "options": {

  787.                     "legend": {

  788.                         "displayMode": "hidden"

  789.                     },

  790.                     "standardOptions": {},

  791.                     "thresholds": {},

  792.                     "tooltip": {

  793.                         "mode": "all",

  794.                         "sort": "none"

  795.                     }

  796.                 },

  797.                 "targets": [

  798.                     {

  799.                         "expr": "topk(5,( max_over_time(namedprocess_namegroup_states{instance=~\"$instance\", groupname=~\"$processes\", state=\"Running\"}[5m]) or max_over_time(namedprocess_namegroup_states{instance=~\"$instance\", groupname=~\"$processes\", state=\"Running\"}[5m])))",

  800.                         "legend": "{{groupname}}",

  801.                         "refId": "A"

  802.                     }

  803.                 ],

  804.                 "type": "timeseries",

  805.                 "version": "2.0.0",

  806.                 "datasourceCate": "prometheus",

  807.                 "datasourceValue": "${prom}"

  808.             },

  809.             {

  810.                 "custom": {

  811.                     "drawStyle": "lines",

  812.                     "fillOpacity": 0.5,

  813.                     "lineInterpolation": "smooth",

  814.                     "stack": "off"

  815.                 },

  816.                 "description": "等待IO状态同名进程数量，倒排前5",

  817.                 "id": "31829466-def5-496f-83e8-10806b19410b",

  818.                 "layout": {

  819.                     "h": 7,

  820.                     "i": "31829466-def5-496f-83e8-10806b19410b",

  821.                     "w": 12,

  822.                     "x": 12,

  823.                     "y": 59

  824.                 },

  825.                 "name": "Top of processes waiting on IO",

  826.                 "options": {

  827.                     "legend": {

  828.                         "displayMode": "hidden"

  829.                     },

  830.                     "standardOptions": {},

  831.                     "thresholds": {},

  832.                     "tooltip": {

  833.                         "mode": "all",

  834.                         "sort": "none"

  835.                     }

  836.                 },

  837.                 "targets": [

  838.                     {

  839.                         "expr": "topk(5,( max_over_time(namedprocess_namegroup_states{instance=~\"$instance\", groupname=~\"$processes\", state=\"Waiting\"}[5m]) or max_over_time(namedprocess_namegroup_states{instance=~\"$instance\", groupname=~\"$processes\", state=\"Waiting\"}[5m])))",

  840.                         "legend": "{{groupname}}",

  841.                         "refId": "A"

  842.                     }

  843.                 ],

  844.                 "type": "timeseries",

  845.                 "version": "2.0.0",

  846.                 "datasourceCate": "prometheus",

  847.                 "datasourceValue": "${prom}"

  848.             },

  849.             {

  850.                 "collapsed": true,

  851.                 "id": "9c79f06d-0262-4284-a65e-bcd0d84130be",

  852.                 "layout": {

  853.                     "h": 1,

  854.                     "i": "9c79f06d-0262-4284-a65e-bcd0d84130be",

  855.                     "w": 24,

  856.                     "x": 0,

  857.                     "y": 66

  858.                 },

  859.                 "name": "Kernel Waits(WCHAN)",

  860.                 "type": "row"

  861.             },

  862.             {

  863.                 "custom": {

  864.                     "drawStyle": "lines",

  865.                     "fillOpacity": 0.5,

  866.                     "lineInterpolation": "smooth",

  867.                     "stack": "off"

  868.                 },

  869.                 "description": "内核函数等待线程数量，倒排前5",

  870.                 "id": "5decb873-5452-4084-8bbf-7bd3c9fa33f4",

  871.                 "layout": {

  872.                     "h": 7,

  873.                     "i": "5decb873-5452-4084-8bbf-7bd3c9fa33f4",

  874.                     "w": 12,

  875.                     "x": 0,

  876.                     "y": 67

  877.                 },

  878.                 "name": "Kernel waits for All",

  879.                 "options": {

  880.                     "legend": {

  881.                         "displayMode": "hidden"

  882.                     },

  883.                     "standardOptions": {},

  884.                     "thresholds": {},

  885.                     "tooltip": {

  886.                         "mode": "all",

  887.                         "sort": "none"

  888.                     }

  889.                 },

  890.                 "targets": [

  891.                     {

  892.                         "expr": "topk(5,sum(avg_over_time(namedprocess_namegroup_threads_wchan{instance=~\"$instance\", groupname=~\"$processes\"}[5m])) by (wchan) )",

  893.                         "legend": "{{wchan}}",

  894.                         "refId": "A"

  895.                     }

  896.                 ],

  897.                 "type": "timeseries",

  898.                 "version": "2.0.0",

  899.                 "datasourceCate": "prometheus",

  900.                 "datasourceValue": "${prom}"

  901.             },

  902.             {

  903.                 "custom": {

  904.                     "drawStyle": "lines",

  905.                     "fillOpacity": 0.5,

  906.                     "lineInterpolation": "smooth",

  907.                     "stack": "off"

  908.                 },

  909.                 "description": "内核函数等待线程数量按进程统计，倒排前5",

  910.                 "id": "20f14e96-5ac8-43f8-93a8-3f7696f5d8d5",

  911.                 "layout": {

  912.                     "h": 7,

  913.                     "i": "20f14e96-5ac8-43f8-93a8-3f7696f5d8d5",

  914.                     "w": 12,

  915.                     "x": 12,

  916.                     "y": 67

  917.                 },

  918.                 "name": "Kernel wait Details for All",

  919.                 "options": {

  920.                     "legend": {

  921.                         "displayMode": "hidden"

  922.                     },

  923.                     "standardOptions": {},

  924.                     "thresholds": {},

  925.                     "tooltip": {

  926.                         "mode": "all",

  927.                         "sort": "none"

  928.                     }

  929.                 },

  930.                 "targets": [

  931.                     {

  932.                         "expr": "topk(5,sum(avg_over_time(namedprocess_namegroup_threads_wchan{instance=~\"$instance\", groupname=~\"$processes\"}[5m])) by (wchan,groupname) )",

  933.                         "legend": "{{wchan}}",

  934.                         "refId": "A"

  935.                     }

  936.                 ],

  937.                 "type": "timeseries",

  938.                 "version": "2.0.0",

  939.                 "datasourceCate": "prometheus",

  940.                 "datasourceValue": "${prom}"

  941.             },

  942.             {

  943.                 "collapsed": true,

  944.                 "id": "972e68da-d03d-4638-81a2-fba73d215788",

  945.                 "layout": {

  946.                     "h": 1,

  947.                     "i": "972e68da-d03d-4638-81a2-fba73d215788",

  948.                     "w": 24,

  949.                     "x": 0,

  950.                     "y": 74

  951.                 },

  952.                 "name": "Uptime",

  953.                 "type": "row"

  954.             },

  955.             {

  956.                 "custom": {

  957.                     "calc": "lastNotNull",

  958.                     "displayMode": "seriesToRows",

  959.                     "showHeader": true

  960.                 },

  961.                 "description": "进程启动时长",

  962.                 "id": "eff1ee11-1d70-4c79-a159-588a363a2b60",

  963.                 "layout": {

  964.                     "h": 7,

  965.                     "i": "eff1ee11-1d70-4c79-a159-588a363a2b60",

  966.                     "w": 24,

  967.                     "x": 0,

  968.                     "y": 75

  969.                 },

  970.                 "name": "Processes by uptime",

  971.                 "options": {

  972.                     "standardOptions": {

  973.                         "util": "seconds"

  974.                     }

  975.                 },

  976.                 "overrides": [

  977.                     {

  978.                         "properties": {

  979.                             "standardOptions": {

  980.                                 "util": "seconds"

  981.                             },

  982.                             "valueMappings": [

  983.                                 {

  984.                                     "match": {

  985.                                         "to": 1800

  986.                                     },

  987.                                     "result": {

  988.                                         "color": "#f91010"

  989.                                     },

  990.                                     "type": "range"

  991.                                 },

  992.                                 {

  993.                                     "match": {

  994.                                         "from": 1800

  995.                                     },

  996.                                     "result": {

  997.                                         "color": "#21f312"

  998.                                     },

  999.                                     "type": "range"

  1000.                                 }

  1001.                             ]

  1002.                         }

  1003.                     }

  1004.                 ],

  1005.                 "targets": [

  1006.                     {

  1007.                         "expr": "time()-(namedprocess_namegroup_oldest_start_time_seconds{instance=~\"$instance\",groupname=~\"$processes\"}>0)",

  1008.                         "legend": "{{groupname}}",

  1009.                         "refId": "A"

  1010.                     }

  1011.                 ],

  1012.                 "type": "table",

  1013.                 "version": "2.0.0",

  1014.                 "datasourceCate": "prometheus",

  1015.                 "datasourceValue": "${prom}"

  1016.             }

  1017.         ],

  1018.         "var": [

  1019.             {

  1020.                 "name": "prom",

  1021.                 "type": "datasource",

  1022.                 "definition": "prometheus"

  1023.             },

  1024.             {

  1025.                 "definition": "label_values(namedprocess_namegroup_num_procs, instance)",

  1026.                 "multi": false,

  1027.                 "name": "instance",

  1028.                 "options": [

  1029.                     "tt-fc-es02.nj:12346"

  1030.                 ],

  1031.                 "type": "query",

  1032.                 "datasource": {

  1033.                     "cate": "prometheus",

  1034.                     "value": "${prom}"

  1035.                 }

  1036.             },

  1037.             {

  1038.                 "allOption": true,

  1039.                 "definition": "label_values(namedprocess_namegroup_cpu_seconds_total{instance=~\"$instance\"},groupname)",

  1040.                 "multi": true,

  1041.                 "name": "processes",

  1042.                 "options": [

  1043.                     "(sd-pam)",

  1044.                     "NetworkManager",

  1045.                     "YDLive",

  1046.                     "YDPython",

  1047.                     "YDService",

  1048.                     "agent",

  1049.                     "agetty",

  1050.                     "atd",

  1051.                     "auditd",

  1052.                     "barad_agent",

  1053.                     "bash",

  1054.                     "chronyd",

  1055.                     "crond",

  1056.                     "dbus-daemon",

  1057.                     "fc-agent",

  1058.                     "fc-alert",

  1059.                     "fc-checker",

  1060.                     "gpg-agent",

  1061.                     "less",

  1062.                     "lsmd",

  1063.                     "mongod",

  1064.                     "mysql",

  1065.                     "mysqld",

  1066.                     "nginx",

  1067.                     "ngo",

  1068.                     "node",

  1069.                     "openvpn",

  1070.                     "podman pause",

  1071.                     "polkitd",

  1072.                     "process-agent",

  1073.                     "redis-server",

  1074.                     "rsyslogd",

  1075.                     "sedispatch",

  1076.                     "sgagent",

  1077.                     "sh",

  1078.                     "ssh-agent",

  1079.                     "sshd",

  1080.                     "sssd",

  1081.                     "sssd_be",

  1082.                     "sssd_nss",

  1083.                     "su",

  1084.                     "sudo",

  1085.                     "systemd",

  1086.                     "systemd-journal",

  1087.                     "systemd-logind",

  1088.                     "systemd-udevd",

  1089.                     "tat_agent",

  1090.                     "trace-agent",

  1091.                     "tuned",

  1092.                     "unbound-anchor",

  1093.                     "vminsert-prod",

  1094.                     "vmselect-prod",

  1095.                     "vmstorage-prod"

  1096.                 ],

  1097.                 "type": "query",

  1098.                 "datasource": {

  1099.                     "cate": "prometheus",

  1100.                     "value": "${prom}"

  1101.                 }

  1102.             }

  1103.         ],

  1104.         "version": "3.0.0"

  1105.     }

  1106. }