Merge pull request #1391 from SomtochiAma/grafana-dash

Update Prometheus labels and dashboard

manifests/monitoring/grafana/dashboards/control-plane.json

@@ -67,7 +67,7 @@
       "pluginVersion": "7.1.1",
       "targets": [
         {
-          "expr": "sum(go_info{kubernetes_namespace=\"$namespace\",kubernetes_pod_name=~\".*-controller-.*\"})",
+          "expr": "sum(go_info{namespace=\"$namespace\",pod=~\".*-controller-.*\"})",
           "interval": "",
           "legendFormat": "pods",
           "refId": "A"
@@ -130,7 +130,7 @@
       "pluginVersion": "7.1.1",
       "targets": [
         {
-          "expr": "max(workqueue_longest_running_processor_seconds{kubernetes_namespace=\"$namespace\",kubernetes_pod_name=~\".*-controller-.*\"})",
+          "expr": "max(workqueue_longest_running_processor_seconds{namespace=\"$namespace\",pod=~\".*-controller-.*\"})",
           "hide": false,
           "interval": "",
           "legendFormat": "seconds",
@@ -192,7 +192,7 @@
       "pluginVersion": "7.1.1",
       "targets": [
         {
-          "expr": "sum(go_memstats_alloc_bytes{kubernetes_namespace=\"$namespace\",kubernetes_pod_name=~\".*-controller-.*\"})",
+          "expr": "sum(go_memstats_alloc_bytes{namespace=\"$namespace\",pod=~\".*-controller-.*\"})",
           "interval": "",
           "legendFormat": "",
           "refId": "A"
@@ -256,7 +256,7 @@
       "pluginVersion": "7.1.1",
       "targets": [
         {
-          "expr": "sum(rate(rest_client_requests_total{kubernetes_namespace=\"$namespace\",kubernetes_pod_name=~\".*-controller-.*\"}[1m]))",
+          "expr": "sum(rate(rest_client_requests_total{namespace=\"$namespace\",pod=~\".*-controller-.*\"}[1m]))",
           "interval": "",
           "legendFormat": "requests",
           "refId": "A"
@@ -329,20 +329,20 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "histogram_quantile(0.50, sum(rate(rest_client_request_latency_seconds_bucket{kubernetes_namespace=\"$namespace\"}[5m])) by (le))",
+          "expr": "histogram_quantile(0.50, sum(rate(rest_client_request_latency_seconds_bucket{namespace=\"$namespace\"}[5m])) by (le))",
           "interval": "",
           "legendFormat": "P50",
           "refId": "A"
         },
         {
-          "expr": "histogram_quantile(0.90, sum(rate(rest_client_request_latency_seconds_bucket{kubernetes_namespace=\"$namespace\"}[5m])) by (le))",
+          "expr": "histogram_quantile(0.90, sum(rate(rest_client_request_latency_seconds_bucket{namespace=\"$namespace\"}[5m])) by (le))",
           "hide": true,
           "interval": "",
           "legendFormat": "P90",
           "refId": "B"
         },
         {
-          "expr": "histogram_quantile(0.99, sum(rate(rest_client_request_latency_seconds_bucket{kubernetes_namespace=\"$namespace\"}[5m])) by (le))",
+          "expr": "histogram_quantile(0.99, sum(rate(rest_client_request_latency_seconds_bucket{namespace=\"$namespace\"}[5m])) by (le))",
           "hide": false,
           "interval": "",
           "legendFormat": "P99",
@@ -441,14 +441,14 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "sum(rate(rest_client_requests_total{kubernetes_namespace=\"$namespace\"}[1m]))",
+          "expr": "sum(rate(rest_client_requests_total{namespace=\"$namespace\"}[1m]))",
           "hide": false,
           "interval": "",
           "legendFormat": "total",
           "refId": "A"
         },
         {
-          "expr": "sum(rate(rest_client_requests_total{kubernetes_namespace=\"$namespace\",code!~\"2..\"}[1m]))",
+          "expr": "sum(rate(rest_client_requests_total{namespace=\"$namespace\",code!~\"2..\"}[1m]))",
           "hide": false,
           "interval": "",
           "legendFormat": "errors",
@@ -545,7 +545,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(process_cpu_seconds_total{kubernetes_namespace=\"$namespace\",kubernetes_pod_name=~\".*-controller-.*\"}[1m])",
+          "expr": "rate(process_cpu_seconds_total{namespace=\"$namespace\",pod=~\".*-controller-.*\"}[1m])",
           "interval": "",
           "legendFormat": "{{kubernetes_pod_name}}",
           "refId": "A"
@@ -640,7 +640,7 @@
       "steppedLine": false,
       "targets": [
         {
-          "expr": "rate(go_memstats_alloc_bytes_total{kubernetes_namespace=\"$namespace\",kubernetes_pod_name=~\".*-controller-.*\"}[1m])",
+          "expr": "rate(go_memstats_alloc_bytes_total{namespace=\"$namespace\",pod=~\".*-controller-.*\"}[1m])",
           "hide": false,
           "interval": "",
           "legendFormat": "{{kubernetes_pod_name}}",

manifests/monitoring/grafana/deployment.yaml

@@ -18,7 +18,7 @@ spec:
     spec:
       containers:
         - name: grafana
-          image: "grafana/grafana:7.2.1"
+          image: "grafana/grafana:7.5.4"
           imagePullPolicy: IfNotPresent
           ports:
             - name: http

manifests/monitoring/prometheus/deployment.yaml

@@ -19,7 +19,7 @@ spec:
       serviceAccountName: prometheus
       containers:
           - name: prometheus
-            image: prom/prometheus:v2.21.0
+            image: prom/prometheus:v2.26.0
             imagePullPolicy: IfNotPresent
             args:
               - '--storage.tsdb.retention=2h'

manifests/monitoring/prometheus/prometheus.yml

@@ -45,8 +45,8 @@ scrape_configs:
   - action: replace
     source_labels:
     - __meta_kubernetes_namespace
-    target_label: kubernetes_namespace
+    target_label: namespace
   - action: replace
     source_labels:
     - __meta_kubernetes_pod_name
-    target_label: kubernetes_pod_name
+    target_label: pod