Fixing mem query, adding vswitchd

Signed-off-by: Joe Talerico aka rook <[email protected]>
cloud-bulldozer · Oct 24, 2024 · 79c1797 · 79c1797
1 parent 9a31f8b
commit 79c1797
Showing 1 changed file with 43 additions and 9 deletions.
diff --git a/pkg/metrics/system.go b/pkg/metrics/system.go
@@ -24,14 +24,16 @@ type NodeInfo struct {
 
 // NodeCPU stores CPU information for a specific Node
 type NodeCPU struct {
-	Idle    float64 `json:"idleCPU"`
-	User    float64 `json:"userCPU"`
-	Steal   float64 `json:"stealCPU"`
-	System  float64 `json:"systemCPU"`
-	Nice    float64 `json:"niceCPU"`
-	Irq     float64 `json:"irqCPU"`
-	Softirq float64 `json:"softCPU"`
-	Iowait  float64 `json:"ioCPU"`
+	Idle       float64 `json:"idleCPU"`
+	User       float64 `json:"userCPU"`
+	Steal      float64 `json:"stealCPU"`
+	System     float64 `json:"systemCPU"`
+	Nice       float64 `json:"niceCPU"`
+	Irq        float64 `json:"irqCPU"`
+	Softirq    float64 `json:"softCPU"`
+	Iowait     float64 `json:"ioCPU"`
+	VSwitchCPU float64 `json:"vSwitchCPU"`
+	VSwitchMem float64 `json:"vSwitchMem"`
 }
 
 // PodCPU stores pod CPU
@@ -193,10 +195,42 @@ func TopPodCPU(node NodeInfo, conn PromConnect, start time.Time, end time.Time)
 	return pods, true
 }
 
+// VSwitchCPU will return the vswitchd cpu usage for specific node
+func VSwitchCPU(node NodeInfo, conn PromConnect, start time.Time, end time.Time, ndata *NodeCPU) bool {
+	query := fmt.Sprintf("irate(container_cpu_usage_seconds_total{id=~\"/system.slice/ovs-vswitchd.service\", node=~\"%s\"}[2m])*100", node.NodeName)
+	logging.Debugf("Prom Query : %s", query)
+	val, err := conn.Client.QueryRange(query, start, end, time.Minute)
+	if err != nil {
+		logging.Error("Issue querying Prometheus")
+		return false
+	}
+	v := val.(model.Matrix)
+	for _, s := range v {
+		ndata.VSwitchCPU = avg(s.Values)
+	}
+	return true
+}
+
+// VSwitchMem will return the vswitchd cpu usage for specific node
+func VSwitchMem(node NodeInfo, conn PromConnect, start time.Time, end time.Time, ndata *NodeCPU) bool {
+	query := fmt.Sprintf("container_memory_rss{id=~\"/system.slice/ovs-vswitchd.service\", node=~\"%s\"}", node.NodeName)
+	logging.Debugf("Prom Query : %s", query)
+	val, err := conn.Client.QueryRange(query, start, end, time.Minute)
+	if err != nil {
+		logging.Error("Issue querying Prometheus")
+		return false
+	}
+	v := val.(model.Matrix)
+	for _, s := range v {
+		ndata.VSwitchMem = avg(s.Values)
+	}
+	return true
+}
+
 // TopPodMem will return the top 5 Mem consumers for a specific node
 func TopPodMem(node NodeInfo, conn PromConnect, start time.Time, end time.Time) (PodValues, bool) {
 	var pods PodValues
-	query := fmt.Sprintf("topk(5,container_memory_rss{container!=\"POD\",name!=\"\",node=~\"%s\"})", node.NodeName)
+	query := fmt.Sprintf("topk(5,sum(container_memory_rss{container!=\"POD\",name!=\"\",node=~\"%s\"})) by (pod, namespace, node))", node.NodeName)
 	logging.Debugf("Prom Query : %s", query)
 	val, err := conn.Client.QueryRange(query, start, end, time.Minute)
 	if err != nil {