123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483 |
- {
- "__inputs": [
- {
- "name": "DS_TELEMETRY-POSTGRES",
- "label": "telemetry-postgres",
- "description": "",
- "type": "datasource",
- "pluginId": "postgres",
- "pluginName": "PostgreSQL"
- }
- ],
- "__elements": [],
- "__requires": [
- {
- "type": "grafana",
- "id": "grafana",
- "name": "Grafana",
- "version": "8.3.2"
- },
- {
- "type": "panel",
- "id": "hpcviz-idvl-hpcc-sankey",
- "name": "sankey",
- "version": "1.0.0"
- },
- {
- "type": "datasource",
- "id": "postgres",
- "name": "PostgreSQL",
- "version": "1.0.0"
- }
- ],
- "editable": false,
- "fiscalYearStartMonth": 0,
- "graphTooltip": 0,
- "id": null,
- "iteration": 1647435013504,
- "links": [],
- "liveNow": false,
- "panels": [
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "gridPos": {
- "h": 19,
- "w": 24,
- "x": 0,
- "y": 0
- },
- "id": 14,
- "options": {
- "coreLimit": 128,
- "displayOpt": "compute_num"
- },
- "targets": [
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "format": "table",
- "group": [],
- "hide": false,
- "metricColumn": "none",
- "panelId": 9,
- "queryType": "randomWalk",
- "rawQuery": true,
- "rawSql": "SELECT $__timeGroupAlias(a.timestamp,$__interval),\na.source_ip, a.jobs, a.cpus \nfrom (\n SELECT timestamp, CONCAT(nodes.servicetag) \n AS nodeid, jobs, cpus, nodes.os_ip_addr AS source_ip\n FROM slurm.node_jobs a \n INNER JOIN nodes \n ON nodes.nodeid = a.nodeid\n WHERE $__timeFilter(a.timestamp)) \nAS a WHERE a.source_ip IN ($NodeByUser)\nGROUP BY a.timestamp, a.source_ip, a.jobs, a.cpus ORDER BY a.timestamp",
- "refId": "node core",
- "select": [
- [
- {
- "params": [
- "value"
- ],
- "type": "column"
- }
- ]
- ],
- "timeColumn": "time",
- "where": [
- {
- "name": "$__timeFilter",
- "params": [],
- "type": "macro"
- }
- ]
- },
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "format": "table",
- "group": [],
- "hide": false,
- "metricColumn": "none",
- "rawQuery": true,
- "rawSql": "SELECT\n *\nFROM\n slurm.jobs\nWHERE\n user_id IN ($Users)\n AND start_time < ${__to:date:seconds}\n AND end_time BETWEEN ${__from:date:seconds} and ${__to:date:seconds}",
- "refId": "jobs",
- "select": [
- [
- {
- "params": [
- "value"
- ],
- "type": "column"
- }
- ]
- ],
- "timeColumn": "time",
- "where": [
- {
- "name": "$__timeFilter",
- "params": [],
- "type": "macro"
- }
- ]
- },
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "format": "time_series",
- "group": [],
- "hide": false,
- "metricColumn": "none",
- "rawQuery": true,
- "rawSql": "SELECT $__timeGroupAlias(\"time\",$__interval),\navg(CAST(value AS decimal)) AS \"memory_power\",\n CONCAT('| ',nodes.os_ip_addr) AS name\nFROM timeseries_metrics\nINNER JOIN nodes\nON nodes.servicetag = timeseries_metrics.system\nWHERE\n$__timeFilter(\"time\") AND\nlabel= 'PowerMetrics TotalMemoryPower' AND\ntimeseries_metrics.system in ($servicetag)\nGROUP BY time,name\nORDER BY time",
- "refId": "memory_power",
- "select": [
- [
- {
- "params": [
- "value"
- ],
- "type": "column"
- }
- ]
- ],
- "timeColumn": "time",
- "where": [
- {
- "name": "$__timeFilter",
- "params": [],
- "type": "macro"
- }
- ]
- },
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "format": "time_series",
- "group": [],
- "hide": false,
- "metricColumn": "none",
- "rawQuery": true,
- "rawSql": "SELECT $__timeGroupAlias(\"time\",$__interval),\navg(CAST(value AS decimal)) AS \"power_consumption\",\n CONCAT('| ',nodes.os_ip_addr) AS name\nFROM timeseries_metrics\nINNER JOIN nodes\nON nodes.servicetag = timeseries_metrics.system\nWHERE\n$__timeFilter(\"time\") AND\nlabel= 'PowerMetrics SystemPowerConsumption' AND\ntimeseries_metrics.system in ($servicetag)\nGROUP BY time,name\nORDER BY time",
- "refId": "power_consumption",
- "select": [
- [
- {
- "params": [
- "value"
- ],
- "type": "column"
- }
- ]
- ],
- "timeColumn": "time",
- "where": [
- {
- "name": "$__timeFilter",
- "params": [],
- "type": "macro"
- }
- ]
- },
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "format": "time_series",
- "group": [],
- "hide": false,
- "metricColumn": "none",
- "rawQuery": true,
- "rawSql": "SELECT $__timeGroupAlias(\"time\",$__interval),\navg(CAST(value AS decimal)) AS \"cpu_power\",\n CONCAT('| ',nodes.os_ip_addr) AS name\nFROM timeseries_metrics\nINNER JOIN nodes\nON nodes.servicetag = timeseries_metrics.system\nWHERE\n$__timeFilter(\"time\") AND\nlabel= 'PowerMetrics TotalCPUPower' AND\ntimeseries_metrics.system in ($servicetag)\nGROUP BY time,name\nORDER BY time",
- "refId": "cpu_power",
- "select": [
- [
- {
- "params": [
- "value"
- ],
- "type": "column"
- }
- ]
- ],
- "timeColumn": "time",
- "where": [
- {
- "name": "$__timeFilter",
- "params": [],
- "type": "macro"
- }
- ]
- },
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "format": "time_series",
- "group": [],
- "hide": false,
- "metricColumn": "none",
- "rawQuery": true,
- "rawSql": "SELECT $__timeGroupAlias(\"time\",$__interval),\navg(CAST(value AS decimal)) AS \"cpu_usage\",\n CONCAT('| ',nodes.os_ip_addr) AS name\nFROM timeseries_metrics\nINNER JOIN nodes\nON nodes.servicetag = timeseries_metrics.system\nWHERE\n$__timeFilter(\"time\") AND\nlabel= 'SystemUsage CPUUsage' AND\ntimeseries_metrics.system in ($servicetag)\nGROUP BY time,name\nORDER BY time",
- "refId": "cpu_usage",
- "select": [
- [
- {
- "params": [
- "value"
- ],
- "type": "column"
- }
- ]
- ],
- "timeColumn": "time",
- "where": [
- {
- "name": "$__timeFilter",
- "params": [],
- "type": "macro"
- }
- ]
- },
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "format": "time_series",
- "group": [],
- "hide": false,
- "metricColumn": "none",
- "rawQuery": true,
- "rawSql": "SELECT $__timeGroupAlias(\"time\",$__interval),\navg(CAST(value AS decimal)) AS \"cpu1_temp\",\n CONCAT('| ',nodes.os_ip_addr) AS name\nFROM timeseries_metrics\nINNER JOIN nodes\nON nodes.servicetag = timeseries_metrics.system\nWHERE\n$__timeFilter(\"time\") AND\nlabel= 'CPU1 Temp TemperatureReading' AND\ntimeseries_metrics.system in ($servicetag) \nGROUP BY time,name\nORDER BY time",
- "refId": "cpu1_temp",
- "select": [
- [
- {
- "params": [
- "value"
- ],
- "type": "column"
- }
- ]
- ],
- "timeColumn": "time",
- "where": [
- {
- "name": "$__timeFilter",
- "params": [],
- "type": "macro"
- }
- ]
- },
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "format": "time_series",
- "group": [],
- "hide": false,
- "metricColumn": "none",
- "rawQuery": true,
- "rawSql": "SELECT $__timeGroupAlias(\"time\",$__interval),\navg(CAST(value AS decimal)) AS \"cpu2_temp\",\n CONCAT('| ',nodes.os_ip_addr) AS name\nFROM timeseries_metrics\nINNER JOIN nodes\nON nodes.servicetag = timeseries_metrics.system\nWHERE\n$__timeFilter(\"time\") AND\nlabel= 'CPU2 Temp TemperatureReading' AND\ntimeseries_metrics.system in ($servicetag) \nGROUP BY time,name\nORDER BY time",
- "refId": "cpu2_temp",
- "select": [
- [
- {
- "params": [
- "value"
- ],
- "type": "column"
- }
- ]
- ],
- "timeColumn": "time",
- "where": [
- {
- "name": "$__timeFilter",
- "params": [],
- "type": "macro"
- }
- ]
- },
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "format": "time_series",
- "group": [],
- "hide": false,
- "metricColumn": "none",
- "rawQuery": true,
- "rawSql": "SELECT $__timeGroupAlias(\"time\",$__interval),\navg(CAST(value AS decimal)) AS \"nic_temp\",\nCONCAT('| ',nodes.os_ip_addr) AS name\nFROM timeseries_metrics\nINNER JOIN nodes\nON nodes.servicetag = timeseries_metrics.system\nWHERE\n$__timeFilter(\"time\") AND\nlabel SIMILAR TO '% NIC 1 Port 1 Partition 1 TemperatureReading' AND\ntimeseries_metrics.system in ($servicetag)\nGROUP BY time,name\nORDER BY time",
- "refId": "nic_temp",
- "select": [
- [
- {
- "params": [
- "value"
- ],
- "type": "column"
- }
- ]
- ],
- "timeColumn": "time",
- "where": [
- {
- "name": "$__timeFilter",
- "params": [],
- "type": "macro"
- }
- ]
- },
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "format": "time_series",
- "group": [],
- "hide": false,
- "metricColumn": "none",
- "rawQuery": true,
- "rawSql": "SELECT $__timeGroupAlias(\"time\",$__interval),\navg(CAST(value AS decimal)) AS \"fan1_speed\",\n CONCAT('| ',nodes.os_ip_addr) AS name\nFROM timeseries_metrics\nINNER JOIN nodes\nON nodes.servicetag = timeseries_metrics.system\nWHERE\n$__timeFilter(\"time\") AND\nlabel= 'Fan 1A RPMReading' AND\ntimeseries_metrics.system in ($servicetag)\nGROUP BY time,name\nORDER BY time",
- "refId": "fan1_speed",
- "select": [
- [
- {
- "params": [
- "value"
- ],
- "type": "column"
- }
- ]
- ],
- "timeColumn": "time",
- "where": [
- {
- "name": "$__timeFilter",
- "params": [],
- "type": "macro"
- }
- ]
- },
- {
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "format": "time_series",
- "group": [],
- "hide": false,
- "metricColumn": "none",
- "rawQuery": true,
- "rawSql": "SELECT $__timeGroupAlias(\"timestamp\",$__interval),\navg(value) AS \"Memory_usage\", \nCONCAT('| ',nodes.os_ip_addr) AS name\nFROM slurm.memoryusage\nINNER JOIN nodes\nON nodes.nodeid = slurm.memoryusage.nodeid\nWHERE\n$__timeFilter(\"timestamp\") AND\nnodes.servicetag in ($servicetag) \nGROUP BY time,name\nORDER BY time",
- "refId": "memory_usage",
- "select": [
- [
- {
- "params": [
- "value"
- ],
- "type": "column"
- }
- ]
- ],
- "timeColumn": "time",
- "where": [
- {
- "name": "$__timeFilter",
- "params": [],
- "type": "macro"
- }
- ]
- }
- ],
- "title": "Sankey",
- "transformations": [],
- "type": "hpcviz-idvl-hpcc-sankey"
- }
- ],
- "refresh": "",
- "schemaVersion": 33,
- "style": "dark",
- "tags": [],
- "templating": {
- "list": [
- {
- "current": {},
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "definition": "SELECT DISTINCT servicetag as __value from nodes",
- "hide": 0,
- "includeAll": true,
- "multi": true,
- "name": "servicetag",
- "options": [],
- "query": "SELECT DISTINCT servicetag as __value from nodes",
- "refresh": 1,
- "regex": "",
- "skipUrlSync": false,
- "sort": 0,
- "type": "query"
- },
- {
- "current": {},
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "definition": "SELECT\n user_id as __value, user_name as __text, nodes as IP, nodes.os_ip_addr, nodes.servicetag\nFROM\n slurm.jobs\nINNER JOIN nodes\nON nodes.os_ip_addr = ANY(nodes)\nWHERE nodes.servicetag in ($servicetag) AND\n start_time < ${__to:date:seconds} AND end_time > ${__from:date:seconds}",
- "hide": 0,
- "includeAll": true,
- "multi": true,
- "name": "Users",
- "options": [],
- "query": "SELECT\n user_id as __value, user_name as __text, nodes as IP, nodes.os_ip_addr, nodes.servicetag\nFROM\n slurm.jobs\nINNER JOIN nodes\nON nodes.os_ip_addr = ANY(nodes)\nWHERE nodes.servicetag in ($servicetag) AND\n start_time < ${__to:date:seconds} AND end_time > ${__from:date:seconds}",
- "refresh": 1,
- "regex": "",
- "skipUrlSync": false,
- "sort": 0,
- "type": "query"
- },
- {
- "current": {},
- "datasource": {
- "type": "postgres",
- "uid": "telemetry-postgres"
- },
- "definition": "SELECT DISTINCT unnest(nodes) as node \nFROM slurm.jobs WHERE \nuser_id IN ($Users) AND start_time < ${__to:date:seconds} AND end_time > ${__from:date:seconds}",
- "hide": 0,
- "includeAll": true,
- "multi": true,
- "name": "NodeByUser",
- "options": [],
- "query": "SELECT DISTINCT unnest(nodes) as node \nFROM slurm.jobs WHERE \nuser_id IN ($Users) AND start_time < ${__to:date:seconds} AND end_time > ${__from:date:seconds}",
- "refresh": 2,
- "regex": "",
- "skipUrlSync": false,
- "sort": 1,
- "type": "query"
- }
- ]
- },
- "time": {
- "from": "now-24h",
- "to": "now"
- },
- "timepicker": {},
- "timezone": "",
- "title": "Sankey",
- "uid": "27YRlmz7y",
- "version": 35,
- "weekStart": ""
- }
|