Small improvements in grafana dashboards.

camilb · Sep 29, 2017 · 90011ee · 90011ee
1 parent cb968ba
commit 90011ee
Show file tree

Hide file tree

Showing 5 changed files with 47 additions and 44 deletions.
diff --git a/operator/assets/grafana/kubernetes-capacity-planing-dashboard.json b/operator/assets/grafana/kubernetes-capacity-planing-dashboard.json
@@ -1043,6 +1043,6 @@
     ]
   },
   "timezone": "browser",
-  "title": "Kubernetes Capacity Planing",
+  "title": "Kubernetes Capacity Planning",
   "version": 4
-}
+}
diff --git a/operator/assets/grafana/kubernetes-cluster-health-dashboard.json b/operator/assets/grafana/kubernetes-cluster-health-dashboard.json
@@ -14,7 +14,7 @@
       "type": "grafana",
       "id": "grafana",
       "name": "Grafana",
-      "version": "4.5.2"
+      "version": "4.4.1"
     },
     {
       "type": "datasource",
@@ -182,7 +182,7 @@
           "tableColumn": "",
           "targets": [
             {
-              "expr": "sum(ALERTS{alertstate=\"firing\"})",
+              "expr": "sum(ALERTS{alertstate=\"firing\",alertname!=\"DeadMansSwitch\"})",
               "format": "time_series",
               "intervalFactor": 2,
               "legendFormat": "",
@@ -197,7 +197,7 @@
           "valueMaps": [
             {
               "op": "=",
-              "text": "N/A",
+              "text": "0",
               "value": "null"
             }
           ],
@@ -260,7 +260,7 @@
           "tableColumn": "",
           "targets": [
             {
-              "expr": "sum(ALERTS{alertstate=\"pending\"})",
+              "expr": "sum(ALERTS{alertstate=\"pending\",alertname!=\"DeadMansSwitch\"})",
               "format": "time_series",
               "intervalFactor": 2,
               "legendFormat": "",
@@ -275,7 +275,7 @@
           "valueMaps": [
             {
               "op": "=",
-              "text": "N/A",
+              "text": "0",
               "value": "null"
             }
           ],
@@ -428,7 +428,7 @@
           "tableColumn": "",
           "targets": [
             {
-              "expr": "sum(kube_node_status_condition{condition!=\"Ready\", status=\"true\"})",
+              "expr": "sum(kube_node_status_condition{condition=\"Ready\",status!=\"true\"})",
               "format": "time_series",
               "intervalFactor": 2,
               "legendFormat": "",
@@ -506,7 +506,7 @@
           "tableColumn": "",
           "targets": [
             {
-              "expr": "sum(kube_node_status_disk_pressure{condition=\"true\"})",
+              "expr": "sum(kube_node_status_condition{condition=\"DiskPressure\",status=\"true\"})",
               "format": "time_series",
               "intervalFactor": 2,
               "legendFormat": "",
@@ -584,7 +584,7 @@
           "tableColumn": "",
           "targets": [
             {
-              "expr": "sum(kube_node_status_memory_pressure{condition=\"true\"})",
+              "expr": "sum(kube_node_status_condition{condition=\"MemoryPressure\",status=\"true\"})",
               "format": "time_series",
               "intervalFactor": 2,
               "legendFormat": "",
@@ -729,5 +729,5 @@
   },
   "timezone": "",
   "title": "Kubernetes Cluster Health",
-  "version": 12
+  "version": 9
 }
diff --git a/operator/assets/grafana/kubernetes-cluster-status-dashboard.json b/operator/assets/grafana/kubernetes-cluster-status-dashboard.json
@@ -176,7 +176,7 @@
           "tableColumn": "",
           "targets": [
             {
-              "expr": "sum(ALERTS{alertstate=\"firing\"})",
+              "expr": "sum(ALERTS{alertstate=\"firing\",alertname!=\"DeadMansSwitch\"})",
               "format": "time_series",
               "intervalFactor": 2,
               "legendFormat": "",
@@ -191,7 +191,7 @@
           "valueMaps": [
             {
               "op": "=",
-              "text": "N/A",
+              "text": "0",
               "value": "null"
             }
           ],
@@ -266,7 +266,7 @@
           "tableColumn": "",
           "targets": [
             {
-              "expr": "(sum(up{job=\"apiserver\"} == 1) / sum(up{job=\"apiserver\"})) * 100",
+              "expr": "(sum(up{job=\"apiserver\"} == 1) / count(up{job=\"apiserver\"})) * 100",
               "format": "time_series",
               "interval": "",
               "intervalFactor": 2,
@@ -345,7 +345,7 @@
           "tableColumn": "",
           "targets": [
             {
-              "expr": "(sum(up{job=\"kube-controller-manager\"} == 1) / sum(up{job=\"kube-controller-manager\"})) * 100",
+              "expr": "(sum(up{job=\"kube-controller-manager\"} == 1) / count(up{job=\"kube-controller-manager\"})) * 100",
               "format": "time_series",
               "interval": "",
               "intervalFactor": 2,
@@ -424,7 +424,7 @@
           "tableColumn": "",
           "targets": [
             {
-              "expr": "(sum(up{job=\"kube-scheduler\"} == 1) / sum(up{job=\"kube-scheduler\"})) * 100",
+              "expr": "(sum(up{job=\"kube-scheduler\"} == 1) / count(up{job=\"kube-scheduler\"})) * 100",
               "format": "time_series",
               "interval": "",
               "intervalFactor": 2,
@@ -892,5 +892,5 @@
   },
   "timezone": "",
   "title": "Kubernetes Cluster Status",
-  "version": 22
-}
+  "version": 3
+}
diff --git a/operator/assets/grafana/kubernetes-control-plane-status-dashboard.json b/operator/assets/grafana/kubernetes-control-plane-status-dashboard.json
@@ -333,7 +333,7 @@
           "tableColumn": "",
           "targets": [
             {
-              "expr": "topk(1, (sum by(instance) (rate(apiserver_request_count{code!~\"2..\"}[5m])) / sum by(instance) (rate(apiserver_request_count[5m]))) * 100)",
+              "expr": "topk(1, (sum by(instance) (rate(apiserver_request_count{code=~\"5..\"}[5m])) / sum by(instance) (rate(apiserver_request_count[5m]))) * 100)",
               "format": "time_series",
               "intervalFactor": 2,
               "legendFormat": "",
@@ -348,7 +348,7 @@
           "valueMaps": [
             {
               "op": "=",
-              "text": "N/A",
+              "text": "0",
               "value": "null"
             }
           ],
@@ -659,5 +659,5 @@
   },
   "timezone": "",
   "title": "Kubernetes Control Plane Status",
-  "version": 11
-}
+  "version": 3
+}
diff --git a/operator/manifests/grafana/grafana-dashboards.cm.yaml b/operator/manifests/grafana/grafana-dashboards.cm.yaml
@@ -1880,9 +1880,10 @@ data:
         ]
       },
       "timezone": "browser",
-      "title": "Kubernetes Capacity Planing",
+      "title": "Kubernetes Capacity Planning",
       "version": 4
-    },
+    }
+    ,
       "inputs": [
         {
           "name": "DS_PROMETHEUS",
@@ -1912,7 +1913,7 @@ data:
           "type": "grafana",
           "id": "grafana",
           "name": "Grafana",
-          "version": "4.5.2"
+          "version": "4.4.1"
         },
         {
           "type": "datasource",
@@ -2080,7 +2081,7 @@ data:
               "tableColumn": "",
               "targets": [
                 {
-                  "expr": "sum(ALERTS{alertstate=\"firing\"})",
+                  "expr": "sum(ALERTS{alertstate=\"firing\",alertname!=\"DeadMansSwitch\"})",
                   "format": "time_series",
                   "intervalFactor": 2,
                   "legendFormat": "",
@@ -2095,7 +2096,7 @@ data:
               "valueMaps": [
                 {
                   "op": "=",
-                  "text": "N/A",
+                  "text": "0",
                   "value": "null"
                 }
               ],
@@ -2158,7 +2159,7 @@ data:
               "tableColumn": "",
               "targets": [
                 {
-                  "expr": "sum(ALERTS{alertstate=\"pending\"})",
+                  "expr": "sum(ALERTS{alertstate=\"pending\",alertname!=\"DeadMansSwitch\"})",
                   "format": "time_series",
                   "intervalFactor": 2,
                   "legendFormat": "",
@@ -2173,7 +2174,7 @@ data:
               "valueMaps": [
                 {
                   "op": "=",
-                  "text": "N/A",
+                  "text": "0",
                   "value": "null"
                 }
               ],
@@ -2326,7 +2327,7 @@ data:
               "tableColumn": "",
               "targets": [
                 {
-                  "expr": "sum(kube_node_status_condition{condition!=\"Ready\", status=\"true\"})",
+                  "expr": "sum(kube_node_status_condition{condition=\"Ready\",status!=\"true\"})",
                   "format": "time_series",
                   "intervalFactor": 2,
                   "legendFormat": "",
@@ -2404,7 +2405,7 @@ data:
               "tableColumn": "",
               "targets": [
                 {
-                  "expr": "sum(kube_node_status_disk_pressure{condition=\"true\"})",
+                  "expr": "sum(kube_node_status_condition{condition=\"DiskPressure\",status=\"true\"})",
                   "format": "time_series",
                   "intervalFactor": 2,
                   "legendFormat": "",
@@ -2482,7 +2483,7 @@ data:
               "tableColumn": "",
               "targets": [
                 {
-                  "expr": "sum(kube_node_status_memory_pressure{condition=\"true\"})",
+                  "expr": "sum(kube_node_status_condition{condition=\"MemoryPressure\",status=\"true\"})",
                   "format": "time_series",
                   "intervalFactor": 2,
                   "legendFormat": "",
@@ -2627,7 +2628,7 @@ data:
       },
       "timezone": "",
       "title": "Kubernetes Cluster Health",
-      "version": 12
+      "version": 9
     }
     ,
       "inputs": [
@@ -3781,7 +3782,7 @@ data:
               "tableColumn": "",
               "targets": [
                 {
-                  "expr": "sum(ALERTS{alertstate=\"firing\"})",
+                  "expr": "sum(ALERTS{alertstate=\"firing\",alertname!=\"DeadMansSwitch\"})",
                   "format": "time_series",
                   "intervalFactor": 2,
                   "legendFormat": "",
@@ -3796,7 +3797,7 @@ data:
               "valueMaps": [
                 {
                   "op": "=",
-                  "text": "N/A",
+                  "text": "0",
                   "value": "null"
                 }
               ],
@@ -3871,7 +3872,7 @@ data:
               "tableColumn": "",
               "targets": [
                 {
-                  "expr": "(sum(up{job=\"apiserver\"} == 1) / sum(up{job=\"apiserver\"})) * 100",
+                  "expr": "(sum(up{job=\"apiserver\"} == 1) / count(up{job=\"apiserver\"})) * 100",
                   "format": "time_series",
                   "interval": "",
                   "intervalFactor": 2,
@@ -3950,7 +3951,7 @@ data:
               "tableColumn": "",
               "targets": [
                 {
-                  "expr": "(sum(up{job=\"kube-controller-manager\"} == 1) / sum(up{job=\"kube-controller-manager\"})) * 100",
+                  "expr": "(sum(up{job=\"kube-controller-manager\"} == 1) / count(up{job=\"kube-controller-manager\"})) * 100",
                   "format": "time_series",
                   "interval": "",
                   "intervalFactor": 2,
@@ -4029,7 +4030,7 @@ data:
               "tableColumn": "",
               "targets": [
                 {
-                  "expr": "(sum(up{job=\"kube-scheduler\"} == 1) / sum(up{job=\"kube-scheduler\"})) * 100",
+                  "expr": "(sum(up{job=\"kube-scheduler\"} == 1) / count(up{job=\"kube-scheduler\"})) * 100",
                   "format": "time_series",
                   "interval": "",
                   "intervalFactor": 2,
@@ -4497,8 +4498,9 @@ data:
       },
       "timezone": "",
       "title": "Kubernetes Cluster Status",
-      "version": 22
-    },
+      "version": 3
+    }
+    ,
       "inputs": [
         {
           "name": "DS_PROMETHEUS",
@@ -4847,7 +4849,7 @@ data:
               "tableColumn": "",
               "targets": [
                 {
-                  "expr": "topk(1, (sum by(instance) (rate(apiserver_request_count{code!~\"2..\"}[5m])) / sum by(instance) (rate(apiserver_request_count[5m]))) * 100)",
+                  "expr": "topk(1, (sum by(instance) (rate(apiserver_request_count{code=~\"5..\"}[5m])) / sum by(instance) (rate(apiserver_request_count[5m]))) * 100)",
                   "format": "time_series",
                   "intervalFactor": 2,
                   "legendFormat": "",
@@ -4862,7 +4864,7 @@ data:
               "valueMaps": [
                 {
                   "op": "=",
-                  "text": "N/A",
+                  "text": "0",
                   "value": "null"
                 }
               ],
@@ -5173,8 +5175,9 @@ data:
       },
       "timezone": "",
       "title": "Kubernetes Control Plane Status",
-      "version": 11
-    },
+      "version": 3
+    }
+    ,
       "inputs": [
         {
           "name": "DS_PROMETHEUS",