2022-02-17 10:13:34 -08:00
local grafana = import 'github.com/grafana/grafonnet-lib/grafonnet/grafana.libsonnet';
local dashboard = grafana.dashboard;
local row = grafana.row;
local prometheus = grafana.prometheus;
local template = grafana.template;
local graphPanel = grafana.graphPanel;
local grafana70 = import 'github.com/grafana/grafonnet-lib/grafonnet-7.0/grafana.libsonnet';
local gaugePanel = grafana70.panel.gauge;
2022-04-27 08:15:50 -07:00
local table = grafana70.panel.table;
2022-02-17 10:13:34 -08:00
{
new(config=null, platform=null):: {
local prometheusDatasourceTemplate = {
current: {
text: 'default',
value: 'default',
},
hide: 0,
label: 'Data Source',
name: 'datasource',
options: [],
query: 'prometheus',
refresh: 1,
regex: '',
type: 'datasource',
},
2024-03-08 05:41:36 -08:00
local clusterTemplatePrototype =
template.new(
'cluster',
'$datasource',
'',
hide=if config.showMultiCluster then '' else '2',
refresh='time',
label='Cluster',
),
local clusterTemplate =
if platform == 'Darwin' then
clusterTemplatePrototype
{ query: 'label_values(node_uname_info{%(nodeExporterSelector)s, sysname="Darwin"}, %(clusterLabel)s)' % config }
else
clusterTemplatePrototype
{ query: 'label_values(node_uname_info{%(nodeExporterSelector)s, sysname!="Darwin"}, %(clusterLabel)s)' % config },
2022-02-17 10:13:34 -08:00
local instanceTemplatePrototype =
template.new(
'instance',
'$datasource',
'',
refresh='time',
label='Instance',
),
local instanceTemplate =
if platform == 'Darwin' then
instanceTemplatePrototype
2024-03-08 05:41:36 -08:00
{ query: 'label_values(node_uname_info{%(nodeExporterSelector)s, %(clusterLabel)s="$cluster", sysname="Darwin"}, instance)' % config }
2022-02-17 10:13:34 -08:00
else
instanceTemplatePrototype
2024-03-08 05:41:36 -08:00
{ query: 'label_values(node_uname_info{%(nodeExporterSelector)s, %(clusterLabel)s="$cluster", sysname!="Darwin"}, instance)' % config },
2022-02-17 10:13:34 -08:00
local idleCPU =
graphPanel.new(
'CPU Usage',
datasource='$datasource',
span=6,
format='percentunit',
max=1,
min=0,
stack=true,
)
.addTarget(prometheus.target(
|||
(
2024-03-08 05:41:36 -08:00
(1 - sum without (mode) (rate(node_cpu_seconds_total{%(nodeExporterSelector)s, mode=~"idle|iowait|steal", instance="$instance", %(clusterLabel)s="$cluster"}[$__rate_interval])))
2022-02-17 10:13:34 -08:00
/ ignoring(cpu) group_left
2024-03-08 05:41:36 -08:00
count without (cpu, mode) (node_cpu_seconds_total{%(nodeExporterSelector)s, mode="idle", instance="$instance", %(clusterLabel)s="$cluster"})
2022-02-17 10:13:34 -08:00
)
||| % config,
legendFormat='{{cpu}}',
intervalFactor=5,
)),
local systemLoad =
graphPanel.new(
'Load Average',
datasource='$datasource',
span=6,
format='short',
min=0,
fill=0,
)
2024-03-08 05:41:36 -08:00
.addTarget(prometheus.target('node_load1{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}' % config, legendFormat='1m load average'))
.addTarget(prometheus.target('node_load5{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}' % config, legendFormat='5m load average'))
.addTarget(prometheus.target('node_load15{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}' % config, legendFormat='15m load average'))
.addTarget(prometheus.target('count(node_cpu_seconds_total{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster", mode="idle"})' % config, legendFormat='logical cores')),
2022-02-17 10:13:34 -08:00
local memoryGraphPanelPrototype =
graphPanel.new(
'Memory Usage',
datasource='$datasource',
span=9,
format='bytes',
min=0,
),
local memoryGraph =
if platform == 'Linux' then
memoryGraphPanelPrototype { stack: true }
.addTarget(prometheus.target(
|||
(
2024-03-08 05:41:36 -08:00
node_memory_MemTotal_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}
2022-02-17 10:13:34 -08:00
-
2024-03-08 05:41:36 -08:00
node_memory_MemFree_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}
2022-02-17 10:13:34 -08:00
-
2024-03-08 05:41:36 -08:00
node_memory_Buffers_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}
2022-02-17 10:13:34 -08:00
-
2024-03-08 05:41:36 -08:00
node_memory_Cached_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}
2022-02-17 10:13:34 -08:00
)
||| % config,
legendFormat='memory used'
))
2024-03-08 05:41:36 -08:00
.addTarget(prometheus.target('node_memory_Buffers_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}' % config, legendFormat='memory buffers'))
.addTarget(prometheus.target('node_memory_Cached_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}' % config, legendFormat='memory cached'))
.addTarget(prometheus.target('node_memory_MemFree_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}' % config, legendFormat='memory free'))
2022-02-17 10:13:34 -08:00
else if platform == 'Darwin' then
// not useful to stack
memoryGraphPanelPrototype { stack: false }
2024-03-08 05:41:36 -08:00
.addTarget(prometheus.target('node_memory_total_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}' % config, legendFormat='Physical Memory'))
2022-02-17 10:13:34 -08:00
.addTarget(prometheus.target(
|||
(
2024-03-08 05:41:36 -08:00
node_memory_internal_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"} -
node_memory_purgeable_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"} +
node_memory_wired_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"} +
node_memory_compressed_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}
2022-02-17 10:13:34 -08:00
)
||| % config, legendFormat='Memory Used'
))
.addTarget(prometheus.target(
|||
(
2024-03-08 05:41:36 -08:00
node_memory_internal_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"} -
node_memory_purgeable_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}
2022-02-17 10:13:34 -08:00
)
||| % config, legendFormat='App Memory'
))
2024-03-08 05:41:36 -08:00
.addTarget(prometheus.target('node_memory_wired_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}' % config, legendFormat='Wired Memory'))
.addTarget(prometheus.target('node_memory_compressed_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}' % config, legendFormat='Compressed')),
2022-02-17 10:13:34 -08:00
// NOTE: avg() is used to circumvent a label change caused by a node_exporter rollout.
local memoryGaugePanelPrototype =
gaugePanel.new(
title='Memory Usage',
datasource='$datasource',
)
.addThresholdStep('rgba(50, 172, 45, 0.97)')
.addThresholdStep('rgba(237, 129, 40, 0.89)', 80)
.addThresholdStep('rgba(245, 54, 54, 0.9)', 90)
.setFieldConfig(max=100, min=0, unit='percent')
+ {
span: 3,
},
local memoryGauge =
if platform == 'Linux' then
memoryGaugePanelPrototype
.addTarget(prometheus.target(
|||
100 -
(
2024-03-08 05:41:36 -08:00
avg(node_memory_MemAvailable_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}) /
avg(node_memory_MemTotal_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"})
2022-02-17 10:13:34 -08:00
* 100
)
||| % config,
))
else if platform == 'Darwin' then
memoryGaugePanelPrototype
.addTarget(prometheus.target(
|||
(
(
2024-03-08 05:41:36 -08:00
avg(node_memory_internal_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}) -
avg(node_memory_purgeable_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}) +
avg(node_memory_wired_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"}) +
avg(node_memory_compressed_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"})
2022-02-17 10:13:34 -08:00
) /
2024-03-08 05:41:36 -08:00
avg(node_memory_total_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster"})
2022-02-17 10:13:34 -08:00
)
*
100
||| % config
)),
local diskIO =
graphPanel.new(
'Disk I/O',
datasource='$datasource',
span=6,
min=0,
fill=0,
)
// TODO: Does it make sense to have those three in the same panel?
.addTarget(prometheus.target(
2024-03-08 05:41:36 -08:00
'rate(node_disk_read_bytes_total{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster", %(diskDeviceSelector)s}[$__rate_interval])' % config,
2022-02-17 10:13:34 -08:00
legendFormat='{{device}} read',
2022-05-12 12:23:31 -07:00
intervalFactor=1,
2022-02-17 10:13:34 -08:00
))
.addTarget(prometheus.target(
2024-03-08 05:41:36 -08:00
'rate(node_disk_written_bytes_total{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster", %(diskDeviceSelector)s}[$__rate_interval])' % config,
2022-02-17 10:13:34 -08:00
legendFormat='{{device}} written',
2022-05-12 12:23:31 -07:00
intervalFactor=1,
2022-02-17 10:13:34 -08:00
))
.addTarget(prometheus.target(
2024-03-08 05:41:36 -08:00
'rate(node_disk_io_time_seconds_total{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster", %(diskDeviceSelector)s}[$__rate_interval])' % config,
2022-02-17 10:13:34 -08:00
legendFormat='{{device}} io time',
2022-05-12 12:23:31 -07:00
intervalFactor=1,
2022-02-17 10:13:34 -08:00
)) +
{
seriesOverrides: [
{
alias: '/ read| written/',
yaxis: 1,
},
{
alias: '/ io time/',
yaxis: 2,
},
],
yaxes: [
2022-05-12 11:38:18 -07:00
self.yaxe(format='Bps'),
2022-05-12 12:23:31 -07:00
self.yaxe(format='percentunit'),
2022-02-17 10:13:34 -08:00
],
},
local diskSpaceUsage =
2022-04-27 08:15:50 -07:00
table.new(
title='Disk Space Usage',
2022-02-17 10:13:34 -08:00
datasource='$datasource',
)
2022-04-27 08:15:50 -07:00
.setFieldConfig(unit='decbytes')
.addThresholdStep(color='green', value=null)
.addThresholdStep(color='yellow', value=0.8)
.addThresholdStep(color='red', value=0.9)
2022-02-17 10:13:34 -08:00
.addTarget(prometheus.target(
|||
2024-03-08 05:41:36 -08:00
max by (mountpoint) (node_filesystem_size_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster", %(fsSelector)s, %(fsMountpointSelector)s})
2022-02-17 10:13:34 -08:00
||| % config,
2022-04-27 08:15:50 -07:00
legendFormat='',
instant=true,
format='table'
2022-02-17 10:13:34 -08:00
))
.addTarget(prometheus.target(
|||
2024-03-08 05:41:36 -08:00
max by (mountpoint) (node_filesystem_avail_bytes{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster", %(fsSelector)s, %(fsMountpointSelector)s})
2022-02-17 10:13:34 -08:00
||| % config,
2022-04-27 08:15:50 -07:00
legendFormat='',
instant=true,
format='table'
))
.addOverride(
matcher={
id: 'byName',
options: 'Mounted on',
},
properties=[
{
id: 'custom.width',
value: 260,
},
],
)
.addOverride(
matcher={
id: 'byName',
options: 'Size',
},
properties=[
{
id: 'custom.width',
value: 93,
},
],
)
.addOverride(
matcher={
id: 'byName',
options: 'Used',
},
properties=[
{
id: 'custom.width',
value: 72,
},
],
)
.addOverride(
matcher={
id: 'byName',
options: 'Available',
},
properties=[
{
id: 'custom.width',
value: 88,
},
],
)
.addOverride(
matcher={
id: 'byName',
options: 'Used, %',
},
properties=[
{
id: 'unit',
value: 'percentunit',
},
{
id: 'custom.displayMode',
value: 'gradient-gauge',
},
{
id: 'max',
value: 1,
},
2022-02-17 10:13:34 -08:00
{
2022-04-27 08:15:50 -07:00
id: 'min',
value: 0,
},
]
)
+ { span: 6 }
+ {
transformations: [
{
id: 'groupBy',
options: {
fields: {
'Value #A': {
aggregations: [
'lastNotNull',
],
operation: 'aggregate',
},
'Value #B': {
aggregations: [
'lastNotNull',
],
operation: 'aggregate',
},
mountpoint: {
aggregations: [],
operation: 'groupby',
},
},
},
2022-02-17 10:13:34 -08:00
},
{
2022-04-27 08:15:50 -07:00
id: 'merge',
options: {},
},
{
id: 'calculateField',
options: {
alias: 'Used',
binary: {
left: 'Value #A (lastNotNull)',
operator: '-',
reducer: 'sum',
right: 'Value #B (lastNotNull)',
},
mode: 'binary',
reduce: {
reducer: 'sum',
},
},
},
{
id: 'calculateField',
options: {
alias: 'Used, %',
binary: {
left: 'Used',
operator: '/',
reducer: 'sum',
right: 'Value #A (lastNotNull)',
},
mode: 'binary',
reduce: {
reducer: 'sum',
},
},
},
{
id: 'organize',
options: {
excludeByName: {},
indexByName: {},
renameByName: {
'Value #A (lastNotNull)': 'Size',
'Value #B (lastNotNull)': 'Available',
mountpoint: 'Mounted on',
},
},
},
{
id: 'sortBy',
options: {
fields: {},
sort: [
{
field: 'Mounted on',
},
],
},
2022-02-17 10:13:34 -08:00
},
],
},
2022-04-27 08:15:50 -07:00
2022-02-17 10:13:34 -08:00
local networkReceived =
graphPanel.new(
'Network Received',
2022-05-12 12:23:31 -07:00
description='Network received (bits/s)',
2022-02-17 10:13:34 -08:00
datasource='$datasource',
span=6,
2022-05-12 11:38:18 -07:00
format='bps',
2022-02-17 10:13:34 -08:00
min=0,
fill=0,
)
.addTarget(prometheus.target(
2024-03-08 05:41:36 -08:00
'rate(node_network_receive_bytes_total{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster", device!="lo"}[$__rate_interval]) * 8' % config,
2022-02-17 10:13:34 -08:00
legendFormat='{{device}}',
2022-05-12 12:23:31 -07:00
intervalFactor=1,
2022-02-17 10:13:34 -08:00
)),
local networkTransmitted =
graphPanel.new(
'Network Transmitted',
2022-05-12 12:23:31 -07:00
description='Network transmitted (bits/s)',
2022-02-17 10:13:34 -08:00
datasource='$datasource',
span=6,
2022-05-12 11:38:18 -07:00
format='bps',
2022-02-17 10:13:34 -08:00
min=0,
fill=0,
)
.addTarget(prometheus.target(
2024-03-08 05:41:36 -08:00
'rate(node_network_transmit_bytes_total{%(nodeExporterSelector)s, instance="$instance", %(clusterLabel)s="$cluster", device!="lo"}[$__rate_interval]) * 8' % config,
2022-02-17 10:13:34 -08:00
legendFormat='{{device}}',
2022-05-12 12:23:31 -07:00
intervalFactor=1,
2022-02-17 10:13:34 -08:00
)),
local cpuRow =
row.new('CPU')
.addPanel(idleCPU)
.addPanel(systemLoad),
local memoryRow =
row.new('Memory')
.addPanel(memoryGraph)
.addPanel(memoryGauge),
local diskRow =
row.new('Disk')
.addPanel(diskIO)
.addPanel(diskSpaceUsage),
local networkRow =
row.new('Network')
.addPanel(networkReceived)
.addPanel(networkTransmitted),
local rows =
[
cpuRow,
memoryRow,
diskRow,
networkRow,
],
local templates =
[
prometheusDatasourceTemplate,
2024-03-08 05:41:36 -08:00
clusterTemplate,
2022-02-17 10:13:34 -08:00
instanceTemplate,
],
dashboard: if platform == 'Linux' then
dashboard.new(
'%sNodes' % config.dashboardNamePrefix,
time_from='now-1h',
tags=(config.dashboardTags),
timezone='utc',
refresh='30s',
graphTooltip='shared_crosshair'
)
.addTemplates(templates)
.addRows(rows)
else if platform == 'Darwin' then
dashboard.new(
'%sMacOS' % config.dashboardNamePrefix,
time_from='now-1h',
tags=(config.dashboardTags),
timezone='utc',
refresh='30s',
graphTooltip='shared_crosshair'
)
.addTemplates(templates)
.addRows(rows),
},
}
