Normalize PVC allocation query

pkg/costmodel/costmodel.go

@@ -217,9 +217,14 @@ const (
 	, "container_name","$1","container","(.+)"), "pod_name","$1","pod","(.+)")`
 	// queryPVCAllocationFmt yields the total byte-hour CPU allocation over the given window.
 	//  sum(all VCPU measurements within given window) = [byte*min] by metric
-	//  (") / 60 = [byte*hour] by metric
+	//  (") / 60 = [byte*hour] by metric, assuming no missed scrapes
+	//  (") * (normalization factor) = [byte*hour] by metric, normalized for missed scrapes
 	//  sum(") by unique pvc = [VCPU*hour] by (cluster, namespace, pod, pv, pvc)
-	queryPVCAllocationFmt     = `sum(sum_over_time(pod_pvc_allocation[%s:1m])) by (cluster_id, namespace, pod, persistentvolume, persistentvolumeclaim) / 60`
+	// Note: normalization factor is 1.0 if no scrapes are missed and has an upper bound determined by minExpectedScrapeRate
+	// so that coarse resolutions don't push normalization factors too high; e.g. 24h resolution with 1h of data would make
+	// for a normalization factor of 24. With a minimumExpectedScrapeRate of 0.95, that caps the norm factor at
+	queryPVCAllocationFmt = `sum(sum_over_time(pod_pvc_allocation[%s:1m])) by (cluster_id, namespace, pod, persistentvolume, persistentvolumeclaim) / 60
+		* 60 / clamp_min(count_over_time(sum(pod_pvc_allocation) by (cluster_id, namespace, pod, persistentvolume, persistentvolumeclaim)[%s:1m])/%f, 60 * %f)`
 	queryPVHourlyCostFmt      = `avg_over_time(pv_hourly_cost[%s])`
 	queryNSLabels             = `avg_over_time(kube_namespace_labels[%s])`
 	queryPodLabels            = `avg_over_time(kube_pod_labels[%s])`
@@ -1698,7 +1703,7 @@ func (cm *CostModel) costDataRange(cli prometheusClient.Client, clientset kubern
 	queryCPUUsage := fmt.Sprintf(queryCPUUsageStr, windowString, "")
 	queryGPURequests := fmt.Sprintf(queryGPURequestsStr, windowString, "", windowString, "", resolutionHours, windowString, "")
 	queryPVRequests := fmt.Sprintf(queryPVRequestsStr)
-	queryPVCAllocation := fmt.Sprintf(queryPVCAllocationFmt, windowString)
+	queryPVCAllocation := fmt.Sprintf(queryPVCAllocationFmt, windowString, windowString, resolutionHours, minimumExpectedScrapeRate)
 	queryPVHourlyCost := fmt.Sprintf(queryPVHourlyCostFmt, windowString)
 	queryNetZoneRequests := fmt.Sprintf(queryZoneNetworkUsage, windowString, "")
 	queryNetRegionRequests := fmt.Sprintf(queryRegionNetworkUsage, windowString, "")