diff --git a/assets/grafana/kubernetes-capacity-planing-dashboard.json b/assets/grafana/kubernetes-capacity-planning-dashboard.json
similarity index 99%
rename from assets/grafana/kubernetes-capacity-planing-dashboard.json
rename to assets/grafana/kubernetes-capacity-planning-dashboard.json
index 8133645301e533231a909291bd18dabe955d31b8..3ea159474ed9931311300946a24e4d15952b2fc0 100644
--- a/assets/grafana/kubernetes-capacity-planing-dashboard.json
+++ b/assets/grafana/kubernetes-capacity-planning-dashboard.json
@@ -1043,6 +1043,6 @@
]
},
"timezone": "browser",
- "title": "Kubernetes Capacity Planing",
+ "title": "Kubernetes Capacity Planning",
"version": 4
}
\ No newline at end of file
diff --git a/assets/grafana/kubernetes-cluster-health-dashboard.json b/assets/grafana/kubernetes-cluster-health-dashboard.json
index d04ff0c0f356f5837603179d7698b457f2701830..46eb6ca7be550861a816af7a9c85c70159ef3e6c 100644
--- a/assets/grafana/kubernetes-cluster-health-dashboard.json
+++ b/assets/grafana/kubernetes-cluster-health-dashboard.json
@@ -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_ready{condition!=\"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
}
\ No newline at end of file
diff --git a/assets/grafana/kubernetes-cluster-status-dashboard.json b/assets/grafana/kubernetes-cluster-status-dashboard.json
index 1b6d5b359e7ba62bec8e5093646cb51a57b11457..d30906bdaf53dd232bcdeada977412c76ed7ecec 100644
--- a/assets/grafana/kubernetes-cluster-status-dashboard.json
+++ b/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
}
\ No newline at end of file
diff --git a/assets/grafana/kubernetes-control-plane-status-dashboard.json b/assets/grafana/kubernetes-control-plane-status-dashboard.json
index 03b547ac3712764ca078e11dfc9d4657e7333f34..9350831334691c2b2ec0a36d4fae919ae3aba005 100644
--- a/assets/grafana/kubernetes-control-plane-status-dashboard.json
+++ b/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
}
\ No newline at end of file
diff --git a/manifests/grafana/grafana-dashboards.yaml b/manifests/grafana/grafana-dashboards.yaml
index 36ac037302e19d982cd55dbca443dc89567b118b..6aec18021bfab7bca5d706d404053d262632c851 100644
--- a/manifests/grafana/grafana-dashboards.yaml
+++ b/manifests/grafana/grafana-dashboards.yaml
@@ -832,7 +832,7 @@ data:
],
"overwrite": true
}
- kubernetes-capacity-planing-dashboard.json: |+
+ kubernetes-capacity-planning-dashboard.json: |+
{
"dashboard":
{
@@ -1880,7 +1880,7 @@ data:
]
},
"timezone": "browser",
- "title": "Kubernetes Capacity Planing",
+ "title": "Kubernetes Capacity Planning",
"version": 4
} ,
"inputs": [
@@ -2080,7 +2080,7 @@ data:
"tableColumn": "",
"targets": [
{
- "expr": "sum(ALERTS{alertstate=\"firing\"})",
+ "expr": "sum(ALERTS{alertstate=\"firing\",alertname!=\"DeadMansSwitch\"})",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "",
@@ -2095,7 +2095,7 @@ data:
"valueMaps": [
{
"op": "=",
- "text": "N/A",
+ "text": "0",
"value": "null"
}
],
@@ -2158,7 +2158,7 @@ data:
"tableColumn": "",
"targets": [
{
- "expr": "sum(ALERTS{alertstate=\"pending\"})",
+ "expr": "sum(ALERTS{alertstate=\"pending\",alertname!=\"DeadMansSwitch\"})",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "",
@@ -2173,7 +2173,7 @@ data:
"valueMaps": [
{
"op": "=",
- "text": "N/A",
+ "text": "0",
"value": "null"
}
],
@@ -2326,7 +2326,7 @@ data:
"tableColumn": "",
"targets": [
{
- "expr": "sum(kube_node_status_ready{condition!=\"true\"})",
+ "expr": "sum(kube_node_status_condition{condition=\"Ready\",status!=\"true\"})",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "",
@@ -2404,7 +2404,7 @@ 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 +2482,7 @@ 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 +2627,7 @@ data:
},
"timezone": "",
"title": "Kubernetes Cluster Health",
- "version": 12
+ "version": 9
} ,
"inputs": [
{
@@ -2820,7 +2820,7 @@ data:
"tableColumn": "",
"targets": [
{
- "expr": "sum(ALERTS{alertstate=\"firing\"})",
+ "expr": "sum(ALERTS{alertstate=\"firing\",alertname!=\"DeadMansSwitch\"})",
"format": "time_series",
"intervalFactor": 2,
"legendFormat": "",
@@ -2835,7 +2835,7 @@ data:
"valueMaps": [
{
"op": "=",
- "text": "N/A",
+ "text": "0",
"value": "null"
}
],
@@ -2910,7 +2910,7 @@ 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,
@@ -2989,7 +2989,7 @@ 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,
@@ -3068,7 +3068,7 @@ 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,
@@ -3536,7 +3536,7 @@ data:
},
"timezone": "",
"title": "Kubernetes Cluster Status",
- "version": 22
+ "version": 3
} ,
"inputs": [
{
@@ -3886,7 +3886,7 @@ 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": "",
@@ -3901,7 +3901,7 @@ data:
"valueMaps": [
{
"op": "=",
- "text": "N/A",
+ "text": "0",
"value": "null"
}
],
@@ -4212,7 +4212,7 @@ data:
},
"timezone": "",
"title": "Kubernetes Control Plane Status",
- "version": 11
+ "version": 3
} ,
"inputs": [
{