Skip to content

Commit f25b9c5

Browse files
sleepyfoodiesleepyfoodie
committed
chore: refactor out queries from rresources/cluster
1 parent 2a29435 commit f25b9c5

File tree

2 files changed

+167
-44
lines changed

2 files changed

+167
-44
lines changed

‎dashboards/resources/cluster.libsonnet‎

Lines changed: 45 additions & 44 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,5 @@
11
local g = import 'github.com/grafana/grafonnet/gen/grafonnet-latest/main.libsonnet';
2+
local queries = import './queries/cluster-queries.libsonnet';
23

34
local prometheus = g.query.prometheus;
45
local stat = g.panel.stat;
@@ -80,47 +81,47 @@ local var = g.dashboard.variable;
8081
statPanel(
8182
'CPU Utilisation',
8283
'percentunit',
83-
'cluster:node_cpu:ratio_rate5m{%(clusterLabel)s="$cluster"}' % $._config
84+
queries.cpuUtilisation($._config)
8485
)
8586
+ stat.gridPos.withW(4)
8687
+ stat.gridPos.withH(3),
8788

8889
statPanel(
8990
'CPU Requests Commitment',
9091
'percentunit',
91-
'sum(namespace_cpu:kube_pod_container_resource_requests:sum{%(clusterLabel)s="$cluster"}) / sum(kube_node_status_allocatable{%(kubeStateMetricsSelector)s,resource="cpu",%(clusterLabel)s="$cluster"})' % $._config
92+
queries.cpuRequestsCommitment($._config)
9293
)
9394
+ stat.gridPos.withW(4)
9495
+ stat.gridPos.withH(3),
9596

9697
statPanel(
9798
'CPU Limits Commitment',
9899
'percentunit',
99-
'sum(namespace_cpu:kube_pod_container_resource_limits:sum{%(clusterLabel)s="$cluster"}) / sum(kube_node_status_allocatable{%(kubeStateMetricsSelector)s,resource="cpu",%(clusterLabel)s="$cluster"})' % $._config
100+
queries.cpuLimitsCommitment($._config)
100101
)
101102
+ stat.gridPos.withW(4)
102103
+ stat.gridPos.withH(3),
103104

104105
statPanel(
105106
'Memory Utilisation',
106107
'percentunit',
107-
'1 - sum(:node_memory_MemAvailable_bytes:sum{%(clusterLabel)s="$cluster"}) / sum(node_memory_MemTotal_bytes{%(nodeExporterSelector)s,%(clusterLabel)s="$cluster"})' % $._config
108+
queries.memoryUtilisation($._config)
108109
)
109110
+ stat.gridPos.withW(4)
110111
+ stat.gridPos.withH(3),
111112

112113
statPanel(
113114
'Memory Requests Commitment',
114115
'percentunit',
115-
'sum(namespace_memory:kube_pod_container_resource_requests:sum{%(clusterLabel)s="$cluster"}) / sum(kube_node_status_allocatable{%(kubeStateMetricsSelector)s,resource="memory",%(clusterLabel)s="$cluster"})' % $._config
116+
queries.memoryRequestsCommitment($._config)
116117
)
117118
+ stat.gridPos.withW(4)
118119
+ stat.gridPos.withH(3),
119120

120121
statPanel(
121122
'Memory Limits Commitment',
122123
'percentunit',
123-
'sum(namespace_memory:kube_pod_container_resource_limits:sum{%(clusterLabel)s="$cluster"}) / sum(kube_node_status_allocatable{%(kubeStateMetricsSelector)s,resource="memory",%(clusterLabel)s="$cluster"})' % $._config
124+
queries.memoryLimitsCommitment($._config)
124125
)
125126
+ stat.gridPos.withW(4)
126127
+ stat.gridPos.withH(3),
@@ -129,38 +130,38 @@ local var = g.dashboard.variable;
129130
+ tsPanel.queryOptions.withTargets([
130131
prometheus.new(
131132
'${datasource}',
132-
'sum(max by (%(clusterLabel)s, %(namespaceLabel)s, pod, container)(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_rate5m{%(clusterLabel)s="$cluster"})) by (namespace)' % $._config
133+
queries.cpuUsageByNamespace($._config)
133134
)
134135
+ prometheus.withLegendFormat('__auto'),
135136
]),
136137

137138
table.new('CPU Quota')
138139
+ table.queryOptions.withTargets([
139-
prometheus.new('${datasource}', 'sum(kube_pod_owner{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster"}) by (namespace)' % $._config)
140+
prometheus.new('${datasource}', queries.podsByNamespace($._config))
140141
+ prometheus.withInstant(true)
141142
+ prometheus.withFormat('table'),
142143

143-
prometheus.new('${datasource}', 'count(avg(namespace_workload_pod:kube_pod_owner:relabel{%(clusterLabel)s="$cluster"}) by (workload, namespace)) by (namespace)' % $._config)
144+
prometheus.new('${datasource}', queries.workloadsByNamespace($._config))
144145
+ prometheus.withInstant(true)
145146
+ prometheus.withFormat('table'),
146147

147-
prometheus.new('${datasource}', 'sum(max by (%(clusterLabel)s, %(namespaceLabel)s, pod, container)(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_rate5m{%(clusterLabel)s="$cluster"})) by (namespace)' % $._config)
148+
prometheus.new('${datasource}', queries.cpuUsageByNamespace($._config))
148149
+ prometheus.withInstant(true)
149150
+ prometheus.withFormat('table'),
150151

151-
prometheus.new('${datasource}', 'sum(namespace_cpu:kube_pod_container_resource_requests:sum{%(clusterLabel)s="$cluster"}) by (namespace)' % $._config)
152+
prometheus.new('${datasource}', queries.cpuRequestsByNamespace($._config))
152153
+ prometheus.withInstant(true)
153154
+ prometheus.withFormat('table'),
154155

155-
prometheus.new('${datasource}', 'sum(max by (%(clusterLabel)s, %(namespaceLabel)s, pod, container)(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_rate5m{%(clusterLabel)s="$cluster"})) by (namespace) / sum(namespace_cpu:kube_pod_container_resource_requests:sum{%(clusterLabel)s="$cluster"}) by (namespace)' % $._config)
156+
prometheus.new('${datasource}', queries.cpuUsageVsRequests($._config))
156157
+ prometheus.withInstant(true)
157158
+ prometheus.withFormat('table'),
158159

159-
prometheus.new('${datasource}', 'sum(namespace_cpu:kube_pod_container_resource_limits:sum{%(clusterLabel)s="$cluster"}) by (namespace)' % $._config)
160+
prometheus.new('${datasource}', queries.cpuLimitsByNamespace($._config))
160161
+ prometheus.withInstant(true)
161162
+ prometheus.withFormat('table'),
162163

163-
prometheus.new('${datasource}', 'sum(max by (%(clusterLabel)s, %(namespaceLabel)s, pod, container)(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_rate5m{%(clusterLabel)s="$cluster"})) by (namespace) / sum(namespace_cpu:kube_pod_container_resource_limits:sum{%(clusterLabel)s="$cluster"}) by (namespace)' % $._config)
164+
prometheus.new('${datasource}', queries.cpuUsageVsLimits($._config))
164165
+ prometheus.withInstant(true)
165166
+ prometheus.withFormat('table'),
166167
])
@@ -246,38 +247,38 @@ local var = g.dashboard.variable;
246247
+ tsPanel.queryOptions.withTargets([
247248
prometheus.new(
248249
'${datasource}',
249-
'sum(max by (%(clusterLabel)s, %(namespaceLabel)s, pod, container)(container_memory_rss{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", container!=""})) by (namespace)' % $._config
250+
queries.memoryUsageByNamespace($._config)
250251
)
251252
+ prometheus.withLegendFormat('__auto'),
252253
]),
253254

254255
table.new('Memory Requests by Namespace')
255256
+ table.queryOptions.withTargets([
256-
prometheus.new('${datasource}', 'sum(kube_pod_owner{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster"}) by (namespace)' % $._config)
257+
prometheus.new('${datasource}', queries.podsByNamespace($._config))
257258
+ prometheus.withInstant(true)
258259
+ prometheus.withFormat('table'),
259260

260-
prometheus.new('${datasource}', 'count(avg(namespace_workload_pod:kube_pod_owner:relabel{%(clusterLabel)s="$cluster"}) by (workload, namespace)) by (namespace)' % $._config)
261+
prometheus.new('${datasource}', queries.workloadsByNamespace($._config))
261262
+ prometheus.withInstant(true)
262263
+ prometheus.withFormat('table'),
263264

264-
prometheus.new('${datasource}', 'sum(max by (%(clusterLabel)s, %(namespaceLabel)s, pod, container)(container_memory_rss{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", container!=""})) by (namespace)' % $._config)
265+
prometheus.new('${datasource}', queries.memoryUsageByNamespace($._config))
265266
+ prometheus.withInstant(true)
266267
+ prometheus.withFormat('table'),
267268

268-
prometheus.new('${datasource}', 'sum(namespace_memory:kube_pod_container_resource_requests:sum{%(clusterLabel)s="$cluster"}) by (namespace)' % $._config)
269+
prometheus.new('${datasource}', queries.memoryRequestsByNamespace($._config))
269270
+ prometheus.withInstant(true)
270271
+ prometheus.withFormat('table'),
271272

272-
prometheus.new('${datasource}', 'sum(max by (%(clusterLabel)s, %(namespaceLabel)s, pod, container)(container_memory_rss{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", container!=""})) by (namespace) / sum(namespace_memory:kube_pod_container_resource_requests:sum{%(clusterLabel)s="$cluster"}) by (namespace)' % $._config)
273+
prometheus.new('${datasource}', queries.memoryUsageVsRequests($._config))
273274
+ prometheus.withInstant(true)
274275
+ prometheus.withFormat('table'),
275276

276-
prometheus.new('${datasource}', 'sum(namespace_memory:kube_pod_container_resource_limits:sum{%(clusterLabel)s="$cluster"}) by (namespace)' % $._config)
277+
prometheus.new('${datasource}', queries.memoryLimitsByNamespace($._config))
277278
+ prometheus.withInstant(true)
278279
+ prometheus.withFormat('table'),
279280

280-
prometheus.new('${datasource}', 'sum(max by (%(clusterLabel)s, %(namespaceLabel)s, pod, container)(container_memory_rss{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", container!=""})) by (namespace) / sum(namespace_memory:kube_pod_container_resource_limits:sum{%(clusterLabel)s="$cluster"}) by (namespace)' % $._config)
281+
prometheus.new('${datasource}', queries.memoryUsageVsLimits($._config))
281282
+ prometheus.withInstant(true)
282283
+ prometheus.withFormat('table'),
283284
])
@@ -396,27 +397,27 @@ local var = g.dashboard.variable;
396397

397398
table.new('Current Network Usage')
398399
+ table.queryOptions.withTargets([
399-
prometheus.new('${datasource}', 'sum(rate(container_network_receive_bytes_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config)
400+
prometheus.new('${datasource}', queries.networkReceiveBandwidth($._config))
400401
+ prometheus.withInstant(true)
401402
+ prometheus.withFormat('table'),
402403

403-
prometheus.new('${datasource}', 'sum(rate(container_network_transmit_bytes_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config)
404+
prometheus.new('${datasource}', queries.networkTransmitBandwidth($._config))
404405
+ prometheus.withInstant(true)
405406
+ prometheus.withFormat('table'),
406407

407-
prometheus.new('${datasource}', 'sum(rate(container_network_receive_packets_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config)
408+
prometheus.new('${datasource}', queries.networkReceivePackets($._config))
408409
+ prometheus.withInstant(true)
409410
+ prometheus.withFormat('table'),
410411

411-
prometheus.new('${datasource}', 'sum(rate(container_network_transmit_packets_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config)
412+
prometheus.new('${datasource}', queries.networkTransmitPackets($._config))
412413
+ prometheus.withInstant(true)
413414
+ prometheus.withFormat('table'),
414415

415-
prometheus.new('${datasource}', 'sum(rate(container_network_receive_packets_dropped_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config)
416+
prometheus.new('${datasource}', queries.networkReceivePacketsDropped($._config))
416417
+ prometheus.withInstant(true)
417418
+ prometheus.withFormat('table'),
418419

419-
prometheus.new('${datasource}', 'sum(rate(container_network_transmit_packets_dropped_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config)
420+
prometheus.new('${datasource}', queries.networkTransmitPacketsDropped($._config))
420421
+ prometheus.withInstant(true)
421422
+ prometheus.withFormat('table'),
422423
])
@@ -510,7 +511,7 @@ local var = g.dashboard.variable;
510511
+ tsPanel.queryOptions.withTargets([
511512
prometheus.new(
512513
'${datasource}',
513-
'sum(rate(container_network_receive_bytes_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config
514+
queries.networkReceiveBandwidth($._config)
514515
)
515516
+ prometheus.withLegendFormat('__auto'),
516517
]),
@@ -520,7 +521,7 @@ local var = g.dashboard.variable;
520521
+ tsPanel.queryOptions.withTargets([
521522
prometheus.new(
522523
'${datasource}',
523-
'sum(rate(container_network_transmit_bytes_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config
524+
queries.networkTransmitBandwidth($._config)
524525
)
525526
+ prometheus.withLegendFormat('__auto'),
526527
]),
@@ -530,7 +531,7 @@ local var = g.dashboard.variable;
530531
+ tsPanel.queryOptions.withTargets([
531532
prometheus.new(
532533
'${datasource}',
533-
'avg(irate(container_network_receive_bytes_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config
534+
queries.avgContainerReceiveBandwidth($._config)
534535
)
535536
+ prometheus.withLegendFormat('__auto'),
536537
]),
@@ -540,7 +541,7 @@ local var = g.dashboard.variable;
540541
+ tsPanel.queryOptions.withTargets([
541542
prometheus.new(
542543
'${datasource}',
543-
'avg(irate(container_network_transmit_bytes_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config
544+
queries.avgContainerTransmitBandwidth($._config)
544545
)
545546
+ prometheus.withLegendFormat('__auto'),
546547
]),
@@ -550,7 +551,7 @@ local var = g.dashboard.variable;
550551
+ tsPanel.queryOptions.withTargets([
551552
prometheus.new(
552553
'${datasource}',
553-
'sum(irate(container_network_receive_packets_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config
554+
queries.rateOfReceivedPackets($._config)
554555
)
555556
+ prometheus.withLegendFormat('__auto'),
556557
]),
@@ -560,7 +561,7 @@ local var = g.dashboard.variable;
560561
+ tsPanel.queryOptions.withTargets([
561562
prometheus.new(
562563
'${datasource}',
563-
'sum(irate(container_network_transmit_packets_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config
564+
queries.rateOfTransmittedPackets($._config)
564565
)
565566
+ prometheus.withLegendFormat('__auto'),
566567
]),
@@ -570,7 +571,7 @@ local var = g.dashboard.variable;
570571
+ tsPanel.queryOptions.withTargets([
571572
prometheus.new(
572573
'${datasource}',
573-
'sum(irate(container_network_receive_packets_dropped_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config
574+
queries.rateOfReceivedPacketsDropped($._config)
574575
)
575576
+ prometheus.withLegendFormat('__auto'),
576577
]),
@@ -580,7 +581,7 @@ local var = g.dashboard.variable;
580581
+ tsPanel.queryOptions.withTargets([
581582
prometheus.new(
582583
'${datasource}',
583-
'sum(irate(container_network_transmit_packets_dropped_total{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", %(namespaceLabel)s=~".+"}[%(grafanaIntervalVar)s])) by (namespace)' % $._config
584+
queries.rateOfTransmittedPacketsDropped($._config)
584585
)
585586
+ prometheus.withLegendFormat('__auto'),
586587
]),
@@ -590,7 +591,7 @@ local var = g.dashboard.variable;
590591
+ tsPanel.queryOptions.withTargets([
591592
prometheus.new(
592593
'${datasource}',
593-
'ceil(sum by(namespace) (rate(container_fs_reads_total{%(cadvisorSelector)s, %(containerfsSelector)s, %(diskDeviceSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s]) + rate(container_fs_writes_total{%(cadvisorSelector)s, %(containerfsSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s])))' % $._config
594+
queries.iopsReadsWrites($._config)
594595
)
595596
+ prometheus.withLegendFormat('__auto'),
596597
]),
@@ -600,34 +601,34 @@ local var = g.dashboard.variable;
600601
+ tsPanel.queryOptions.withTargets([
601602
prometheus.new(
602603
'${datasource}',
603-
'sum by(namespace) (rate(container_fs_reads_bytes_total{%(cadvisorSelector)s, %(containerfsSelector)s, %(diskDeviceSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s]) + rate(container_fs_writes_bytes_total{%(cadvisorSelector)s, %(containerfsSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s]))' % $._config
604+
queries.throughputReadWrite($._config)
604605
)
605606
+ prometheus.withLegendFormat('__auto'),
606607
]),
607608

608609
table.new('Current Storage IO')
609610
+ table.queryOptions.withTargets([
610-
prometheus.new('${datasource}', 'sum by(namespace) (rate(container_fs_reads_total{%(cadvisorSelector)s, %(diskDeviceSelector)s, %(containerfsSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s]))' % $._config)
611+
prometheus.new('${datasource}', queries.iopsReads($._config))
611612
+ prometheus.withInstant(true)
612613
+ prometheus.withFormat('table'),
613614

614-
prometheus.new('${datasource}', 'sum by(namespace) (rate(container_fs_writes_total{%(cadvisorSelector)s, %(diskDeviceSelector)s, %(containerfsSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s]))' % $._config)
615+
prometheus.new('${datasource}', queries.iopsWrites($._config))
615616
+ prometheus.withInstant(true)
616617
+ prometheus.withFormat('table'),
617618

618-
prometheus.new('${datasource}', 'sum by(namespace) (rate(container_fs_reads_total{%(cadvisorSelector)s, %(diskDeviceSelector)s, %(containerfsSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s]) + rate(container_fs_writes_total{%(cadvisorSelector)s, %(diskDeviceSelector)s, %(containerfsSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s]))' % $._config)
619+
prometheus.new('${datasource}', queries.iopsReadsWritesCombined($._config))
619620
+ prometheus.withInstant(true)
620621
+ prometheus.withFormat('table'),
621622

622-
prometheus.new('${datasource}', 'sum by(namespace) (rate(container_fs_reads_bytes_total{%(cadvisorSelector)s, %(diskDeviceSelector)s, %(containerfsSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s]))' % $._config)
623+
prometheus.new('${datasource}', queries.throughputRead($._config))
623624
+ prometheus.withInstant(true)
624625
+ prometheus.withFormat('table'),
625626

626-
prometheus.new('${datasource}', 'sum by(namespace) (rate(container_fs_writes_bytes_total{%(cadvisorSelector)s, %(diskDeviceSelector)s, %(containerfsSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s]))' % $._config)
627+
prometheus.new('${datasource}', queries.throughputWrite($._config))
627628
+ prometheus.withInstant(true)
628629
+ prometheus.withFormat('table'),
629630

630-
prometheus.new('${datasource}', 'sum by(namespace) (rate(container_fs_reads_bytes_total{%(cadvisorSelector)s, %(diskDeviceSelector)s, %(containerfsSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s]) + rate(container_fs_writes_bytes_total{%(cadvisorSelector)s, %(diskDeviceSelector)s, %(containerfsSelector)s, %(clusterLabel)s="$cluster", namespace!=""}[%(grafanaIntervalVar)s]))' % $._config)
631+
prometheus.new('${datasource}', queries.throughputReadWriteCombined($._config))
631632
+ prometheus.withInstant(true)
632633
+ prometheus.withFormat('table'),
633634
])

0 commit comments

Comments
 (0)