Просмотр исходного кода

Merge pull request #2293 from r2k1/sum_over_time

Simplify prometheus queries
Cliff Colvin 2 лет назад
Родитель
Сommit
901cdf620f
1 измененных файлов с 15 добавлено и 31 удалено
  1. 15 31
      pkg/costmodel/costmodel.go

+ 15 - 31
pkg/costmodel/costmodel.go

@@ -144,39 +144,25 @@ const (
 		label_replace(
 			label_replace(
 				avg(
-					count_over_time(kube_pod_container_resource_requests{resource="memory", unit="byte", container!="",container!="POD", node!="", %s}[%s] %s)
-					*
-					avg_over_time(kube_pod_container_resource_requests{resource="memory", unit="byte", container!="",container!="POD", node!="", %s}[%s] %s)
+					sum_over_time(kube_pod_container_resource_requests{resource="memory", unit="byte", container!="",container!="POD", node!="", %s}[%s] %s)
 				) by (namespace,container,pod,node,%s) , "container_name","$1","container","(.+)"
 			), "pod_name","$1","pod","(.+)"
 		)
 	) by (namespace,container_name,pod_name,node,%s)`
-	queryRAMUsageStr = `sort_desc(
-		avg(
-			label_replace(
-				label_replace(
-					label_replace(
-						count_over_time(container_memory_working_set_bytes{container!="", container!="POD", instance!="", %s}[%s] %s), "node", "$1", "instance", "(.+)"
-					), "container_name", "$1", "container", "(.+)"
-				), "pod_name", "$1", "pod", "(.+)"
-			)
-			*
+	queryRAMUsageStr = `avg(
+		label_replace(
 			label_replace(
 				label_replace(
-					label_replace(
-						avg_over_time(container_memory_working_set_bytes{container!="", container!="POD", instance!="", %s}[%s] %s), "node", "$1", "instance", "(.+)"
-					), "container_name", "$1", "container", "(.+)"
-				), "pod_name", "$1", "pod", "(.+)"
-			)
-		) by (namespace, container_name, pod_name, node, %s)
-	)`
+					sum_over_time(container_memory_working_set_bytes{container!="", container!="POD", instance!="", %s}[%s] %s), "node", "$1", "instance", "(.+)"
+				), "container_name", "$1", "container", "(.+)"
+			), "pod_name", "$1", "pod", "(.+)"
+		)
+	) by (namespace, container_name, pod_name, node, %s)`
 	queryCPURequestsStr = `avg(
 		label_replace(
 			label_replace(
 				avg(
-					count_over_time(kube_pod_container_resource_requests{resource="cpu", unit="core", container!="",container!="POD", node!="", %s}[%s] %s)
-					*
-					avg_over_time(kube_pod_container_resource_requests{resource="cpu", unit="core", container!="",container!="POD", node!="", %s}[%s] %s)
+					sum_over_time(kube_pod_container_resource_requests{resource="cpu", unit="core", container!="",container!="POD", node!="", %s}[%s] %s)
 				) by (namespace,container,pod,node,%s) , "container_name","$1","container","(.+)"
 			), "pod_name","$1","pod","(.+)"
 		)
@@ -196,9 +182,7 @@ const (
 		label_replace(
 			label_replace(
 				avg(
-					count_over_time(kube_pod_container_resource_requests{resource="nvidia_com_gpu", container!="",container!="POD", node!="", %s}[%s] %s)
-					*
-					avg_over_time(kube_pod_container_resource_requests{resource="nvidia_com_gpu", container!="",container!="POD", node!="", %s}[%s] %s)
+					sum_over_time(kube_pod_container_resource_requests{resource="nvidia_com_gpu", container!="",container!="POD", node!="", %s}[%s] %s)
 					* %f
 				) by (namespace,container,pod,node,%s) , "container_name","$1","container","(.+)"
 			), "pod_name","$1","pod","(.+)"
@@ -253,7 +237,7 @@ const (
 )
 
 func (cm *CostModel) ComputeCostData(cli prometheusClient.Client, cp costAnalyzerCloud.Provider, window string, offset string, filterNamespace string) (map[string]*CostData, error) {
-	queryRAMUsage := fmt.Sprintf(queryRAMUsageStr, env.GetPromClusterFilter(), window, offset, env.GetPromClusterFilter(), window, offset, env.GetPromClusterLabel())
+	queryRAMUsage := fmt.Sprintf(queryRAMUsageStr, env.GetPromClusterFilter(), window, offset, env.GetPromClusterLabel())
 	queryCPUUsage := fmt.Sprintf(queryCPUUsageStr, env.GetPromClusterFilter(), window, offset, env.GetPromClusterLabel())
 	queryNetZoneRequests := fmt.Sprintf(queryZoneNetworkUsage, env.GetPromClusterFilter(), window, "", env.GetPromClusterLabel())
 	queryNetRegionRequests := fmt.Sprintf(queryRegionNetworkUsage, env.GetPromClusterFilter(), window, "", env.GetPromClusterLabel())
@@ -1699,11 +1683,11 @@ func (cm *CostModel) costDataRange(cli prometheusClient.Client, cp costAnalyzerC
 
 	queryRAMAlloc := fmt.Sprintf(queryRAMAllocationByteHours, env.GetPromClusterFilter(), resStr, env.GetPromClusterLabel(), scrapeIntervalSeconds)
 	queryCPUAlloc := fmt.Sprintf(queryCPUAllocationVCPUHours, env.GetPromClusterFilter(), resStr, env.GetPromClusterLabel(), scrapeIntervalSeconds)
-	queryRAMRequests := fmt.Sprintf(queryRAMRequestsStr, env.GetPromClusterFilter(), resStr, "", env.GetPromClusterFilter(), resStr, "", env.GetPromClusterLabel(), env.GetPromClusterLabel())
-	queryRAMUsage := fmt.Sprintf(queryRAMUsageStr, env.GetPromClusterFilter(), resStr, "", env.GetPromClusterFilter(), resStr, "", env.GetPromClusterLabel())
-	queryCPURequests := fmt.Sprintf(queryCPURequestsStr, env.GetPromClusterFilter(), resStr, "", env.GetPromClusterFilter(), resStr, "", env.GetPromClusterLabel(), env.GetPromClusterLabel())
+	queryRAMRequests := fmt.Sprintf(queryRAMRequestsStr, env.GetPromClusterFilter(), resStr, "", env.GetPromClusterLabel(), env.GetPromClusterLabel())
+	queryRAMUsage := fmt.Sprintf(queryRAMUsageStr, env.GetPromClusterFilter(), resStr, "", env.GetPromClusterLabel())
+	queryCPURequests := fmt.Sprintf(queryCPURequestsStr, env.GetPromClusterFilter(), resStr, "", env.GetPromClusterLabel(), env.GetPromClusterLabel())
 	queryCPUUsage := fmt.Sprintf(queryCPUUsageStr, env.GetPromClusterFilter(), resStr, "", env.GetPromClusterLabel())
-	queryGPURequests := fmt.Sprintf(queryGPURequestsStr, env.GetPromClusterFilter(), resStr, "", env.GetPromClusterFilter(), resStr, "", resolution.Hours(), env.GetPromClusterLabel(), env.GetPromClusterLabel(), env.GetPromClusterLabel(), env.GetPromClusterFilter(), resStr, "", env.GetPromClusterLabel())
+	queryGPURequests := fmt.Sprintf(queryGPURequestsStr, env.GetPromClusterFilter(), resStr, "", resolution.Hours(), env.GetPromClusterLabel(), env.GetPromClusterLabel(), env.GetPromClusterLabel(), env.GetPromClusterFilter(), resStr, "", env.GetPromClusterLabel())
 	queryPVRequests := fmt.Sprintf(queryPVRequestsStr, env.GetPromClusterFilter(), env.GetPromClusterLabel(), env.GetPromClusterLabel(), env.GetPromClusterFilter(), env.GetPromClusterLabel(), env.GetPromClusterLabel())
 	queryPVCAllocation := fmt.Sprintf(queryPVCAllocationFmt, env.GetPromClusterFilter(), resStr, env.GetPromClusterLabel(), scrapeIntervalSeconds)
 	queryPVHourlyCost := fmt.Sprintf(queryPVHourlyCostFmt, env.GetPromClusterFilter(), resStr)