fixup! Use kube-scheduler's metrics instead of kube-state-metrics

rexagod · rexagod · commit 758d714b2af5 · 2023-01-11T14:57:36.000+05:30
diff --git a/dashboards/resources/multi-cluster.libsonnet b/dashboards/resources/multi-cluster.libsonnet
@@ -26,23 +26,23 @@ local g = import 'github.com/grafana/jsonnet-libs/grafana-builder/grafana.libson
           )
           .addPanel(
             g.panel('CPU Requests Commitment') +
-            g.statPanel('sum(kube_pod_resource_requests{%(kubeStateMetricsSelector)s, resource="cpu"}) / sum(kube_node_status_allocatable{%(kubeStateMetricsSelector)s, resource="cpu"})' % $._config)
+            g.statPanel('sum(kube_pod_resource_requests{%(kubeSchedulerSelector)s, resource="cpu"}) / sum(kube_node_status_allocatable{%(kubeSchedulerSelector)s, resource="cpu"})' % $._config)
           )
           .addPanel(
             g.panel('CPU Limits Commitment') +
-            g.statPanel('sum(kube_pod_resource_limits{%(kubeStateMetricsSelector)s, resource="cpu"}) / sum(kube_node_status_allocatable{%(kubeStateMetricsSelector)s, resource="cpu"})' % $._config)
+            g.statPanel('sum(kube_pod_resource_limits{%(kubeSchedulerSelector)s, resource="cpu"}) / sum(kube_node_status_allocatable{%(kubeSchedulerSelector)s, resource="cpu"})' % $._config)
           )
           .addPanel(
             g.panel('Memory Utilisation') +
             g.statPanel('1 - sum(:node_memory_MemAvailable_bytes:sum) / sum(node_memory_MemTotal_bytes{%(nodeExporterSelector)s})' % $._config)
           )
           .addPanel(
             g.panel('Memory Requests Commitment') +
-            g.statPanel('sum(kube_pod_resource_requests{%(kubeStateMetricsSelector)s, resource="memory"}) / sum(kube_node_status_allocatable{%(kubeStateMetricsSelector)s, resource="memory"})' % $._config)
+            g.statPanel('sum(kube_pod_resource_requests{%(kubeSchedulerSelector)s, resource="memory"}) / sum(kube_node_status_allocatable{%(kubeSchedulerSelector)s, resource="memory"})' % $._config)
           )
           .addPanel(
             g.panel('Memory Limits Commitment') +
-            g.statPanel('sum(kube_pod_resource_limits{%(kubeStateMetricsSelector)s, resource="memory"}) / sum(kube_node_status_allocatable{%(kubeStateMetricsSelector)s, resource="memory"})' % $._config)
+            g.statPanel('sum(kube_pod_resource_limits{%(kubeSchedulerSelector)s, resource="memory"}) / sum(kube_node_status_allocatable{%(kubeSchedulerSelector)s, resource="memory"})' % $._config)
           )
         )
         .addRow(
@@ -59,10 +59,10 @@ local g = import 'github.com/grafana/jsonnet-libs/grafana-builder/grafana.libson
             g.panel('CPU Quota') +
             g.tablePanel([
               'sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate) by (%(clusterLabel)s)' % $._config,
-              'sum(kube_pod_resource_requests{%(kubeStateMetricsSelector)s, resource="cpu"}) by (%(clusterLabel)s)' % $._config,
-              'sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate) by (%(clusterLabel)s) / sum(kube_pod_resource_requests{%(kubeStateMetricsSelector)s, resource="cpu"}) by (%(clusterLabel)s)' % $._config,
-              'sum(kube_pod_resource_limits{%(kubeStateMetricsSelector)s, resource="cpu"}) by (%(clusterLabel)s)' % $._config,
-              'sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate) by (%(clusterLabel)s) / sum(kube_pod_resource_limits{%(kubeStateMetricsSelector)s, resource="cpu"}) by (%(clusterLabel)s)' % $._config,
+              'sum(kube_pod_resource_requests{%(kubeSchedulerSelector)s, resource="cpu"}) by (%(clusterLabel)s)' % $._config,
+              'sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate) by (%(clusterLabel)s) / sum(kube_pod_resource_requests{%(kubeSchedulerSelector)s, resource="cpu"}) by (%(clusterLabel)s)' % $._config,
+              'sum(kube_pod_resource_limits{%(kubeSchedulerSelector)s, resource="cpu"}) by (%(clusterLabel)s)' % $._config,
+              'sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate) by (%(clusterLabel)s) / sum(kube_pod_resource_limits{%(kubeSchedulerSelector)s, resource="cpu"}) by (%(clusterLabel)s)' % $._config,
             ], tableStyles {
               'Value #A': { alias: 'CPU Usage' },
               'Value #B': { alias: 'CPU Requests' },
@@ -88,10 +88,10 @@ local g = import 'github.com/grafana/jsonnet-libs/grafana-builder/grafana.libson
             g.tablePanel([
               // Not using container_memory_usage_bytes here because that includes page cache
               'sum(container_memory_rss{%(cadvisorSelector)s, container!=""}) by (%(clusterLabel)s)' % $._config,
-              'sum(kube_pod_resource_requests{%(kubeStateMetricsSelector)s, resource="memory"}) by (%(clusterLabel)s)' % $._config,
-              'sum(container_memory_rss{%(cadvisorSelector)s, container!=""}) by (%(clusterLabel)s) / sum(kube_pod_resource_requests{%(kubeStateMetricsSelector)s, resource="memory"}) by (%(clusterLabel)s)' % $._config,
-              'sum(kube_pod_resource_limits{%(kubeStateMetricsSelector)s, resource="memory"}) by (%(clusterLabel)s)' % $._config,
-              'sum(container_memory_rss{%(cadvisorSelector)s, container!=""}) by (%(clusterLabel)s) / sum(kube_pod_resource_limits{%(kubeStateMetricsSelector)s, resource="memory"}) by (%(clusterLabel)s)' % $._config,
+              'sum(kube_pod_resource_requests{%(kubeSchedulerSelector)s, resource="memory"}) by (%(clusterLabel)s)' % $._config,
+              'sum(container_memory_rss{%(cadvisorSelector)s, container!=""}) by (%(clusterLabel)s) / sum(kube_pod_resource_requests{%(kubeSchedulerSelector)s, resource="memory"}) by (%(clusterLabel)s)' % $._config,
+              'sum(kube_pod_resource_limits{%(kubeSchedulerSelector)s, resource="memory"}) by (%(clusterLabel)s)' % $._config,
+              'sum(container_memory_rss{%(cadvisorSelector)s, container!=""}) by (%(clusterLabel)s) / sum(kube_pod_resource_limits{%(kubeSchedulerSelector)s, resource="memory"}) by (%(clusterLabel)s)' % $._config,
             ], tableStyles {
               'Value #A': { alias: 'Memory Usage', unit: 'bytes' },
               'Value #B': { alias: 'Memory Requests', unit: 'bytes' },
diff --git a/dashboards/resources/namespace.libsonnet b/dashboards/resources/namespace.libsonnet
@@ -141,19 +141,19 @@ local template = grafana.template;
          })
         .addPanel(
           g.panel('CPU Utilisation (from requests)') +
-          g.statPanel('sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{%(clusterLabel)s="$cluster", namespace="$namespace"}) / sum(kube_pod_resource_requests{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"})' % $._config)
+          g.statPanel('sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{%(clusterLabel)s="$cluster", namespace="$namespace"}) / sum(kube_pod_resource_requests{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"})' % $._config)
         )
         .addPanel(
           g.panel('CPU Utilisation (from limits)') +
-          g.statPanel('sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{%(clusterLabel)s="$cluster", namespace="$namespace"}) / sum(kube_pod_resource_limits{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"})' % $._config)
+          g.statPanel('sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{%(clusterLabel)s="$cluster", namespace="$namespace"}) / sum(kube_pod_resource_limits{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"})' % $._config)
         )
         .addPanel(
           g.panel('Memory Utilisation (from requests)') +
-          g.statPanel('sum(container_memory_working_set_bytes{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace",container!="", image!=""}) / sum(kube_pod_resource_requests{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="memory"})' % $._config)
+          g.statPanel('sum(container_memory_working_set_bytes{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace",container!="", image!=""}) / sum(kube_pod_resource_requests{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="memory"})' % $._config)
         )
         .addPanel(
           g.panel('Memory Utilisation (from limits)') +
-          g.statPanel('sum(container_memory_working_set_bytes{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace",container!="", image!=""}) / sum(kube_pod_resource_limits{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="memory"})' % $._config)
+          g.statPanel('sum(container_memory_working_set_bytes{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace",container!="", image!=""}) / sum(kube_pod_resource_limits{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="memory"})' % $._config)
         )
       )
       .addRow(
diff --git a/dashboards/resources/pod.libsonnet b/dashboards/resources/pod.libsonnet
@@ -50,7 +50,7 @@ local template = grafana.template;
 
       local cpuRequestsQuery = |||
         sum(
-            kube_pod_resource_requests{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", pod="$pod", resource="cpu"}
+            kube_pod_resource_requests{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", pod="$pod", resource="cpu"}
         )
       ||| % $._config;
 
diff --git a/dashboards/resources/workload-namespace.libsonnet b/dashboards/resources/workload-namespace.libsonnet
@@ -137,7 +137,7 @@ local template = grafana.template;
 
       local cpuRequestsQuery = |||
         sum(
-          kube_pod_resource_requests{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"}
+          kube_pod_resource_requests{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"}
         * on(namespace,pod)
           group_left(workload, workload_type) namespace_workload_pod:kube_pod_owner:relabel{%(clusterLabel)s="$cluster", namespace="$namespace", workload_type="$type"}
         ) by (workload, workload_type)
diff --git a/dashboards/resources/workload.libsonnet b/dashboards/resources/workload.libsonnet
@@ -135,7 +135,7 @@ local template = grafana.template;
 
       local cpuRequestsQuery = |||
         sum(
-            kube_pod_resource_requests{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"}
+            kube_pod_resource_requests{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"}
           * on(namespace,pod)
             group_left(workload, workload_type) namespace_workload_pod:kube_pod_owner:relabel{%(clusterLabel)s="$cluster", namespace="$namespace", workload="$workload", workload_type="$type"}
         ) by (pod)
diff --git a/rules/apps.libsonnet b/rules/apps.libsonnet
@@ -61,7 +61,7 @@
           {
             record: 'cluster:namespace:pod_memory:active:kube_pod_resource_requests',
             expr: |||
-              kube_pod_resource_requests{resource="memory",%(kubeStateMetricsSelector)s}  * on (namespace, pod, %(clusterLabel)s)
+              kube_pod_resource_requests{resource="memory",%(kubeSchedulerSelector)s}  * on (namespace, pod, %(clusterLabel)s)
               group_left() max by (namespace, pod, %(clusterLabel)s) (
                 (kube_pod_status_phase{phase=~"Pending|Running"} == 1)
               )
@@ -73,7 +73,7 @@
               sum by (namespace, %(clusterLabel)s) (
                   sum by (namespace, pod, %(clusterLabel)s) (
                       max by (namespace, pod, container, %(clusterLabel)s) (
-                        kube_pod_resource_requests{resource="memory",%(kubeStateMetricsSelector)s}
+                        kube_pod_resource_requests{resource="memory",%(kubeSchedulerSelector)s}
                       ) * on(namespace, pod, %(clusterLabel)s) group_left() max by (namespace, pod, %(clusterLabel)s) (
                         kube_pod_status_phase{phase=~"Pending|Running"} == 1
                       )
@@ -84,7 +84,7 @@
           {
             record: 'cluster:namespace:pod_cpu:active:kube_pod_resource_requests',
             expr: |||
-              kube_pod_resource_requests{resource="cpu",%(kubeStateMetricsSelector)s}  * on (namespace, pod, %(clusterLabel)s)
+              kube_pod_resource_requests{resource="cpu",%(kubeSchedulerSelector)s}  * on (namespace, pod, %(clusterLabel)s)
               group_left() max by (namespace, pod, %(clusterLabel)s) (
                 (kube_pod_status_phase{phase=~"Pending|Running"} == 1)
               )
@@ -96,7 +96,7 @@
               sum by (namespace, %(clusterLabel)s) (
                   sum by (namespace, pod, %(clusterLabel)s) (
                       max by (namespace, pod, container, %(clusterLabel)s) (
-                        kube_pod_resource_requests{resource="cpu",%(kubeStateMetricsSelector)s}
+                        kube_pod_resource_requests{resource="cpu",%(kubeSchedulerSelector)s}
                       ) * on(namespace, pod, %(clusterLabel)s) group_left() max by (namespace, pod, %(clusterLabel)s) (
                         kube_pod_status_phase{phase=~"Pending|Running"} == 1
                       )
@@ -107,7 +107,7 @@
           {
             record: 'cluster:namespace:pod_memory:active:kube_pod_resource_limits',
             expr: |||
-              kube_pod_resource_limits{resource="memory",%(kubeStateMetricsSelector)s}  * on (namespace, pod, %(clusterLabel)s)
+              kube_pod_resource_limits{resource="memory",%(kubeSchedulerSelector)s}  * on (namespace, pod, %(clusterLabel)s)
               group_left() max by (namespace, pod, %(clusterLabel)s) (
                 (kube_pod_status_phase{phase=~"Pending|Running"} == 1)
               )
@@ -119,7 +119,7 @@
               sum by (namespace, %(clusterLabel)s) (
                   sum by (namespace, pod, %(clusterLabel)s) (
                       max by (namespace, pod, container, %(clusterLabel)s) (
-                        kube_pod_resource_limits{resource="memory",%(kubeStateMetricsSelector)s}
+                        kube_pod_resource_limits{resource="memory",%(kubeSchedulerSelector)s}
                       ) * on(namespace, pod, %(clusterLabel)s) group_left() max by (namespace, pod, %(clusterLabel)s) (
                         kube_pod_status_phase{phase=~"Pending|Running"} == 1
                       )
@@ -130,7 +130,7 @@
           {
             record: 'cluster:namespace:pod_cpu:active:kube_pod_resource_limits',
             expr: |||
-              kube_pod_resource_limits{resource="cpu",%(kubeStateMetricsSelector)s}  * on (namespace, pod, %(clusterLabel)s)
+              kube_pod_resource_limits{resource="cpu",%(kubeSchedulerSelector)s}  * on (namespace, pod, %(clusterLabel)s)
               group_left() max by (namespace, pod, %(clusterLabel)s) (
                (kube_pod_status_phase{phase=~"Pending|Running"} == 1)
                )
@@ -142,7 +142,7 @@
               sum by (namespace, %(clusterLabel)s) (
                   sum by (namespace, pod, %(clusterLabel)s) (
                       max by (namespace, pod, container, %(clusterLabel)s) (
-                        kube_pod_resource_limits{resource="cpu",%(kubeStateMetricsSelector)s}
+                        kube_pod_resource_limits{resource="cpu",%(kubeSchedulerSelector)s}
                       ) * on(namespace, pod, %(clusterLabel)s) group_left() max by (namespace, pod, %(clusterLabel)s) (
                         kube_pod_status_phase{phase=~"Pending|Running"} == 1
                       )
diff --git a/rules/windows.libsonnet b/rules/windows.libsonnet
@@ -217,28 +217,28 @@
             record: 'kube_pod_windows_container_resource_memory_request',
             expr: |||
               max by (namespace, pod, container) (
-                kube_pod_resource_requests{resource="memory",%(kubeStateMetricsSelector)s}
+                kube_pod_resource_requests{resource="memory",%(kubeSchedulerSelector)s}
               ) * on(container,pod,namespace) (windows_pod_container_available)
             ||| % $._config,
           },
           {
             record: 'kube_pod_windows_container_resource_memory_limit',
             expr: |||
-              kube_pod_resource_limits{resource="memory",%(kubeStateMetricsSelector)s} * on(container,pod,namespace) (windows_pod_container_available)
+              kube_pod_resource_limits{resource="memory",%(kubeSchedulerSelector)s} * on(container,pod,namespace) (windows_pod_container_available)
             ||| % $._config,
           },
           {
             record: 'kube_pod_windows_container_resource_cpu_cores_request',
             expr: |||
               max by (namespace, pod, container) (
-                kube_pod_resource_requests{resource="cpu",%(kubeStateMetricsSelector)s}
+                kube_pod_resource_requests{resource="cpu",%(kubeSchedulerSelector)s}
               ) * on(container,pod,namespace) (windows_pod_container_available)
             ||| % $._config,
           },
           {
             record: 'kube_pod_windows_container_resource_cpu_cores_limit',
             expr: |||
-              kube_pod_resource_limits{resource="cpu",%(kubeStateMetricsSelector)s} * on(container,pod,namespace) (windows_pod_container_available)
+              kube_pod_resource_limits{resource="cpu",%(kubeSchedulerSelector)s} * on(container,pod,namespace) (windows_pod_container_available)
             ||| % $._config,
           },
           {

Original file line number	Diff line number	Diff line change
`@@ -141,19 +141,19 @@ local template = grafana.template;`
`141`	`141`	`})`
`142`	`142`	`.addPanel(`
`143`	`143`	`g.panel('CPU Utilisation (from requests)') +`
`144`		`- g.statPanel('sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{%(clusterLabel)s="$cluster", namespace="$namespace"}) / sum(kube_pod_resource_requests{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"})' % $._config)`
	`144`	`+ g.statPanel('sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{%(clusterLabel)s="$cluster", namespace="$namespace"}) / sum(kube_pod_resource_requests{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"})' % $._config)`
`145`	`145`	`)`
`146`	`146`	`.addPanel(`
`147`	`147`	`g.panel('CPU Utilisation (from limits)') +`
`148`		`- g.statPanel('sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{%(clusterLabel)s="$cluster", namespace="$namespace"}) / sum(kube_pod_resource_limits{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"})' % $._config)`
	`148`	`+ g.statPanel('sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{%(clusterLabel)s="$cluster", namespace="$namespace"}) / sum(kube_pod_resource_limits{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="cpu"})' % $._config)`
`149`	`149`	`)`
`150`	`150`	`.addPanel(`
`151`	`151`	`g.panel('Memory Utilisation (from requests)') +`
`152`		`- g.statPanel('sum(container_memory_working_set_bytes{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace",container!="", image!=""}) / sum(kube_pod_resource_requests{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="memory"})' % $._config)`
	`152`	`+ g.statPanel('sum(container_memory_working_set_bytes{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace",container!="", image!=""}) / sum(kube_pod_resource_requests{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="memory"})' % $._config)`
`153`	`153`	`)`
`154`	`154`	`.addPanel(`
`155`	`155`	`g.panel('Memory Utilisation (from limits)') +`
`156`		`- g.statPanel('sum(container_memory_working_set_bytes{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace",container!="", image!=""}) / sum(kube_pod_resource_limits{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="memory"})' % $._config)`
	`156`	`+ g.statPanel('sum(container_memory_working_set_bytes{%(cadvisorSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace",container!="", image!=""}) / sum(kube_pod_resource_limits{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", resource="memory"})' % $._config)`
`157`	`157`	`)`
`158`	`158`	`)`
`159`	`159`	`.addRow(`
Original file line number	Diff line number	Diff line change
`@@ -50,7 +50,7 @@ local template = grafana.template;`
`50`	`50`
`51`	`51`	`local cpuRequestsQuery = \|\|\|`
`52`	`52`	`sum(`
`53`		`- kube_pod_resource_requests{%(kubeStateMetricsSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", pod="$pod", resource="cpu"}`
	`53`	`+ kube_pod_resource_requests{%(kubeSchedulerSelector)s, %(clusterLabel)s="$cluster", namespace="$namespace", pod="$pod", resource="cpu"}`
`54`	`54`	`)`
`55`	`55`	`\|\|\| % $._config;`
`56`	`56`