Prometheu: add more metrics (#800)

* feat(platform,monitor): add more metrics for k8s node Signed-off-by: Feng Kun <[email protected]> * feat(platform,monitor): add more k8s cluster metrics Signed-off-by: Feng Kun <[email protected]>
tkestack · Oct 11, 2020 · ebab1d7 · ebab1d7
1 parent 2e08b9d
commit ebab1d7
Show file tree

Hide file tree

Showing 2 changed files with 72 additions and 0 deletions.
diff --git a/pkg/monitor/controller/prometheus/yamls.go b/pkg/monitor/controller/prometheus/yamls.go
@@ -1062,6 +1062,42 @@ groups:
  - record: k8s_cluster_node_num
  expr: count(kube_node_info)
 
+ - record: k8s_cluster_node_not_ready_num
+ expr: count(k8s_node_status_ready) - count(k8s_node_status_ready == 1)
+
+ - record: k8s_cluster_node_not_ready_rate
+ expr: k8s_cluster_node_not_ready_num / count(k8s_node_status_ready) * 100
+
+ - record: k8s_cluster_pod_status_phase_failed_num
+ expr: count(kube_pod_status_phase{phase="Failed"} == 1)
+
+ - record: k8s_cluster_pod_status_phase_failed_rate
+ expr: count(kube_pod_status_phase{phase="Failed"} == 1) / count(kube_pod_status_phase{phase="Failed"}) * 100
+
+ - record: k8s_cluster_pod_status_phase_pending_num
+ expr: count(kube_pod_status_phase{phase="Pending"} == 1)
+
+ - record: k8s_cluster_pod_status_phase_pending_rate
+ expr: count(kube_pod_status_phase{phase="Pending"} == 1) / count(kube_pod_status_phase{phase="Pending"}) * 100
+
+ - record: k8s_cluster_pod_status_phase_running_num
+ expr: count(kube_pod_status_phase{phase="Running"} == 1)
+
+ - record: k8s_cluster_pod_status_phase_running_rate
+ expr: count(kube_pod_status_phase{phase="Running"} == 1) / count(kube_pod_status_phase{phase="Running"}) * 100
+
+ - record: k8s_cluster_pod_status_phase_succeeded_num
+ expr: count(kube_pod_status_phase{phase="Succeeded"} == 1)
+
+ - record: k8s_cluster_pod_status_phase_succeeded_rate
+ expr: count(kube_pod_status_phase{phase="Succeeded"} == 1) / count(kube_pod_status_phase{phase="Succeeded"}) * 100
+
+ - record: k8s_cluster_pod_status_phase_unknown_num
+ expr: count(kube_pod_status_phase{phase="Unknown"} == 1)
+
+ - record: k8s_cluster_pod_status_phase_unknown_rate
+ expr: count(kube_pod_status_phase{phase="Unknown"} == 1) / count(kube_pod_status_phase{phase="Unknown"}) * 100
+
  - record: k8s_cluster_pod_num
  expr: sum(k8s_node_pod_num)
 

diff --git a/pkg/platform/controller/addon/prometheus/yamls.go b/pkg/platform/controller/addon/prometheus/yamls.go
@@ -1062,6 +1062,42 @@ groups:
  - record: k8s_cluster_node_num
  expr: count(kube_node_info)
 
+ - record: k8s_cluster_node_not_ready_num
+ expr: count(k8s_node_status_ready) - count(k8s_node_status_ready == 1)
+
+ - record: k8s_cluster_node_not_ready_rate
+ expr: k8s_cluster_node_not_ready_num / count(k8s_node_status_ready) * 100
+
+ - record: k8s_cluster_pod_status_phase_failed_num
+ expr: count(kube_pod_status_phase{phase="Failed"} == 1)
+
+ - record: k8s_cluster_pod_status_phase_failed_rate
+ expr: count(kube_pod_status_phase{phase="Failed"} == 1) / count(kube_pod_status_phase{phase="Failed"}) * 100
+
+ - record: k8s_cluster_pod_status_phase_pending_num
+ expr: count(kube_pod_status_phase{phase="Pending"} == 1)
+
+ - record: k8s_cluster_pod_status_phase_pending_rate
+ expr: count(kube_pod_status_phase{phase="Pending"} == 1) / count(kube_pod_status_phase{phase="Pending"}) * 100
+
+ - record: k8s_cluster_pod_status_phase_running_num
+ expr: count(kube_pod_status_phase{phase="Running"} == 1)
+
+ - record: k8s_cluster_pod_status_phase_running_rate
+ expr: count(kube_pod_status_phase{phase="Running"} == 1) / count(kube_pod_status_phase{phase="Running"}) * 100
+
+ - record: k8s_cluster_pod_status_phase_succeeded_num
+ expr: count(kube_pod_status_phase{phase="Succeeded"} == 1)
+
+ - record: k8s_cluster_pod_status_phase_succeeded_rate
+ expr: count(kube_pod_status_phase{phase="Succeeded"} == 1) / count(kube_pod_status_phase{phase="Succeeded"}) * 100
+
+ - record: k8s_cluster_pod_status_phase_unknown_num
+ expr: count(kube_pod_status_phase{phase="Unknown"} == 1)
+
+ - record: k8s_cluster_pod_status_phase_unknown_rate
+ expr: count(kube_pod_status_phase{phase="Unknown"} == 1) / count(kube_pod_status_phase{phase="Unknown"}) * 100
+
  - record: k8s_cluster_pod_num
  expr: sum(k8s_node_pod_num)