cost-model/pkg/costmodel/cluster.go

package costmodel

import (
	"fmt"
	"net"
	"strconv"
	"strings"
	"time"

	"github.com/opencost/opencost/pkg/cloud/provider"
	prometheus "github.com/prometheus/client_golang/api"
	"golang.org/x/exp/slices"

	"github.com/opencost/opencost/core/pkg/log"
	"github.com/opencost/opencost/core/pkg/opencost"
	"github.com/opencost/opencost/core/pkg/util/timeutil"
	"github.com/opencost/opencost/pkg/cloud/models"
	"github.com/opencost/opencost/pkg/env"
	"github.com/opencost/opencost/pkg/prom"
)

const (
	queryClusterCores = `sum(
		avg(avg_over_time(kube_node_status_capacity_cpu_cores{%s}[%s] %s)) by (node, %s) * avg(avg_over_time(node_cpu_hourly_cost{%s}[%s] %s)) by (node, %s) * 730 +
		avg(avg_over_time(node_gpu_hourly_cost{%s}[%s] %s)) by (node, %s) * 730
	  ) by (%s)`

	queryClusterRAM = `sum(
		avg(avg_over_time(kube_node_status_capacity_memory_bytes{%s}[%s] %s)) by (node, %s) / 1024 / 1024 / 1024 * avg(avg_over_time(node_ram_hourly_cost{%s}[%s] %s)) by (node, %s) * 730
	  ) by (%s)`

	queryStorage = `sum(
		avg(avg_over_time(pv_hourly_cost{%s}[%s] %s)) by (persistentvolume, %s) * 730
		* avg(avg_over_time(kube_persistentvolume_capacity_bytes{%s}[%s] %s)) by (persistentvolume, %s) / 1024 / 1024 / 1024
	  ) by (%s) %s`

	queryTotal = `sum(avg(node_total_hourly_cost{%s}) by (node, %s)) * 730 +
	  sum(
		avg(avg_over_time(pv_hourly_cost{%s}[1h])) by (persistentvolume, %s) * 730
		* avg(avg_over_time(kube_persistentvolume_capacity_bytes{%s}[1h])) by (persistentvolume, %s) / 1024 / 1024 / 1024
	  ) by (%s) %s`

	queryNodes = `sum(avg(node_total_hourly_cost{%s}) by (node, %s)) * 730 %s`
)

const MAX_LOCAL_STORAGE_SIZE = 1024 * 1024 * 1024 * 1024

// When ASSET_INCLUDE_LOCAL_DISK_COST is set to false, local storage
// provisioned by sig-storage-local-static-provisioner is excluded
// by checking if the volume is prefixed by "local-pv-".
//
// This is based on the sig-storage-local-static-provisioner implementation,
// which creates all PVs with the "local-pv-" prefix. For reference, see:
// https://github.com/kubernetes-sigs/sig-storage-local-static-provisioner/blob/b6f465027bd059e92c0032c81dd1e1d90e35c909/pkg/discovery/discovery.go#L410-L417
const SIG_STORAGE_LOCAL_PROVISIONER_PREFIX = "local-pv-"

// Costs represents cumulative and monthly cluster costs over a given duration. Costs
// are broken down by cores, memory, and storage.
type ClusterCosts struct {
	Start             *time.Time             `json:"startTime"`
	End               *time.Time             `json:"endTime"`
	CPUCumulative     float64                `json:"cpuCumulativeCost"`
	CPUMonthly        float64                `json:"cpuMonthlyCost"`
	CPUBreakdown      *ClusterCostsBreakdown `json:"cpuBreakdown"`
	GPUCumulative     float64                `json:"gpuCumulativeCost"`
	GPUMonthly        float64                `json:"gpuMonthlyCost"`
	RAMCumulative     float64                `json:"ramCumulativeCost"`
	RAMMonthly        float64                `json:"ramMonthlyCost"`
	RAMBreakdown      *ClusterCostsBreakdown `json:"ramBreakdown"`
	StorageCumulative float64                `json:"storageCumulativeCost"`
	StorageMonthly    float64                `json:"storageMonthlyCost"`
	StorageBreakdown  *ClusterCostsBreakdown `json:"storageBreakdown"`
	TotalCumulative   float64                `json:"totalCumulativeCost"`
	TotalMonthly      float64                `json:"totalMonthlyCost"`
	DataMinutes       float64
}

// ClusterCostsBreakdown provides percentage-based breakdown of a resource by
// categories: user for user-space (i.e. non-system) usage, system, and idle.
type ClusterCostsBreakdown struct {
	Idle   float64 `json:"idle"`
	Other  float64 `json:"other"`
	System float64 `json:"system"`
	User   float64 `json:"user"`
}

// NewClusterCostsFromCumulative takes cumulative cost data over a given time range, computes
// the associated monthly rate data, and returns the Costs.
func NewClusterCostsFromCumulative(cpu, gpu, ram, storage float64, window, offset time.Duration, dataHours float64) (*ClusterCosts, error) {
	start, end := timeutil.ParseTimeRange(window, offset)

	// If the number of hours is not given (i.e. is zero) compute one from the window and offset
	if dataHours == 0 {
		dataHours = end.Sub(start).Hours()
	}

	// Do not allow zero-length windows to prevent divide-by-zero issues
	if dataHours == 0 {
		return nil, fmt.Errorf("illegal time range: window %s, offset %s", window, offset)
	}

	cc := &ClusterCosts{
		Start:             &start,
		End:               &end,
		CPUCumulative:     cpu,
		GPUCumulative:     gpu,
		RAMCumulative:     ram,
		StorageCumulative: storage,
		TotalCumulative:   cpu + gpu + ram + storage,
		CPUMonthly:        cpu / dataHours * (timeutil.HoursPerMonth),
		GPUMonthly:        gpu / dataHours * (timeutil.HoursPerMonth),
		RAMMonthly:        ram / dataHours * (timeutil.HoursPerMonth),
		StorageMonthly:    storage / dataHours * (timeutil.HoursPerMonth),
	}
	cc.TotalMonthly = cc.CPUMonthly + cc.GPUMonthly + cc.RAMMonthly + cc.StorageMonthly

	return cc, nil
}

type Disk struct {
	Cluster        string
	Name           string
	ProviderID     string
	StorageClass   string
	VolumeName     string
	ClaimName      string
	ClaimNamespace string
	Cost           float64
	Bytes          float64

	// These two fields may not be available at all times because they rely on
	// a new set of metrics that may or may not be available. Thus, they must
	// be nilable to represent the complete absence of the data.
	//
	// In other words, nilability here lets us distinguish between
	// "metric is not available" and "metric is available but is 0".
	//
	// They end in "Ptr" to distinguish from an earlier version in order to
	// ensure that all usages are checked for nil.
	BytesUsedAvgPtr *float64
	BytesUsedMaxPtr *float64

	Local     bool
	Start     time.Time
	End       time.Time
	Minutes   float64
	Breakdown *ClusterCostsBreakdown
}

type DiskIdentifier struct {
	Cluster string
	Name    string
}

func ClusterDisks(client prometheus.Client, cp models.Provider, start, end time.Time) (map[DiskIdentifier]*Disk, error) {
	// Start from the time "end", querying backwards
	t := end

	// minsPerResolution determines accuracy and resource use for the following
	// queries. Smaller values (higher resolution) result in better accuracy,
	// but more expensive queries, and vice-a-versa.
	resolution := env.GetETLResolution()
	//Ensuring if ETL_RESOLUTION_SECONDS is less than 60s default it to 1m
	var minsPerResolution int
	if minsPerResolution = int(resolution.Minutes()); int(resolution.Minutes()) == 0 {
		minsPerResolution = 1
		log.DedupedWarningf(3, "ClusterDisks(): Configured ETL resolution (%d seconds) is below the 60 seconds threshold. Overriding with 1 minute.", int(resolution.Seconds()))
	}

	durStr := timeutil.DurationString(end.Sub(start))
	if durStr == "" {
		return nil, fmt.Errorf("illegal duration value for %s", opencost.NewClosedWindow(start, end))
	}

	ctx := prom.NewNamedContext(client, prom.ClusterContextName)
	queryPVCost := fmt.Sprintf(`avg(avg_over_time(pv_hourly_cost{%s}[%s])) by (%s, persistentvolume,provider_id)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryPVSize := fmt.Sprintf(`avg(avg_over_time(kube_persistentvolume_capacity_bytes{%s}[%s])) by (%s, persistentvolume)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryActiveMins := fmt.Sprintf(`avg(kube_persistentvolume_capacity_bytes{%s}) by (%s, persistentvolume)[%s:%dm]`, env.GetPromClusterFilter(), env.GetPromClusterLabel(), durStr, minsPerResolution)
	queryPVStorageClass := fmt.Sprintf(`avg(avg_over_time(kubecost_pv_info{%s}[%s])) by (%s, persistentvolume, storageclass)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryPVUsedAvg := fmt.Sprintf(`avg(avg_over_time(kubelet_volume_stats_used_bytes{%s}[%s])) by (%s, persistentvolumeclaim, namespace)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryPVUsedMax := fmt.Sprintf(`max(max_over_time(kubelet_volume_stats_used_bytes{%s}[%s])) by (%s, persistentvolumeclaim, namespace)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryPVCInfo := fmt.Sprintf(`avg(avg_over_time(kube_persistentvolumeclaim_info{%s}[%s])) by (%s, volumename, persistentvolumeclaim, namespace)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())

	resChPVCost := ctx.QueryAtTime(queryPVCost, t)
	resChPVSize := ctx.QueryAtTime(queryPVSize, t)
	resChActiveMins := ctx.QueryAtTime(queryActiveMins, t)
	resChPVStorageClass := ctx.QueryAtTime(queryPVStorageClass, t)
	resChPVUsedAvg := ctx.QueryAtTime(queryPVUsedAvg, t)
	resChPVUsedMax := ctx.QueryAtTime(queryPVUsedMax, t)
	resChPVCInfo := ctx.QueryAtTime(queryPVCInfo, t)

	resPVCost, _ := resChPVCost.Await()
	resPVSize, _ := resChPVSize.Await()
	resActiveMins, _ := resChActiveMins.Await()
	resPVStorageClass, _ := resChPVStorageClass.Await()
	resPVUsedAvg, _ := resChPVUsedAvg.Await()
	resPVUsedMax, _ := resChPVUsedMax.Await()
	resPVCInfo, _ := resChPVCInfo.Await()

	// Cloud providers do not always charge for a node's local disk costs (i.e.
	// ephemeral storage). Provide an option to opt out of calculating &
	// allocating local disk costs. Note, that this does not affect
	// PersistentVolume costs.
	//
	// Ref:
	// https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/RootDeviceStorage.html
	// https://learn.microsoft.com/en-us/azure/virtual-machines/managed-disks-overview#temporary-disk
	// https://cloud.google.com/compute/docs/disks/local-ssd
	resLocalStorageCost := []*prom.QueryResult{}
	resLocalStorageUsedCost := []*prom.QueryResult{}
	resLocalStorageUsedAvg := []*prom.QueryResult{}
	resLocalStorageUsedMax := []*prom.QueryResult{}
	resLocalStorageBytes := []*prom.QueryResult{}
	resLocalActiveMins := []*prom.QueryResult{}
	if env.GetAssetIncludeLocalDiskCost() {
		// hourlyToCumulative is a scaling factor that, when multiplied by an
		// hourly value, converts it to a cumulative value; i.e. [$/hr] *
		// [min/res]*[hr/min] = [$/res]
		hourlyToCumulative := float64(minsPerResolution) * (1.0 / 60.0)
		costPerGBHr := 0.04 / 730.0

		// container_fs metrics contains metrics for disks that are not local storage of the node. While not perfect to
		// attempt to identify the correct device which is being used as local storage we first filter for devices mounted
		// at paths `/dev/nvme.*` or `/dev/sda.*`. There still may be multiple devices mounted at paths matching the regex
		// so later on we will select the device with the highest `container_fs_limit_bytes` per instance to create a local disk asset
		queryLocalStorageCost := fmt.Sprintf(`sum_over_time(sum(container_fs_limit_bytes{device=~"/dev/(nvme|sda).*", id="/", %s}) by (instance, device, %s)[%s:%dm]) / 1024 / 1024 / 1024 * %f * %f`, env.GetPromClusterFilter(), env.GetPromClusterLabel(), durStr, minsPerResolution, hourlyToCumulative, costPerGBHr)
		queryLocalStorageUsedCost := fmt.Sprintf(`sum_over_time(sum(container_fs_usage_bytes{device=~"/dev/(nvme|sda).*", id="/", %s}) by (instance, device, %s)[%s:%dm]) / 1024 / 1024 / 1024 * %f * %f`, env.GetPromClusterFilter(), env.GetPromClusterLabel(), durStr, minsPerResolution, hourlyToCumulative, costPerGBHr)
		queryLocalStorageUsedAvg := fmt.Sprintf(`avg(sum(avg_over_time(container_fs_usage_bytes{device=~"/dev/(nvme|sda).*", id="/", %s}[%s])) by (instance, device, %s, job)) by (instance, device, %s)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel(), env.GetPromClusterLabel())
		queryLocalStorageUsedMax := fmt.Sprintf(`max(sum(max_over_time(container_fs_usage_bytes{device=~"/dev/(nvme|sda).*", id="/", %s}[%s])) by (instance, device, %s, job)) by (instance, device, %s)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel(), env.GetPromClusterLabel())
		queryLocalStorageBytes := fmt.Sprintf(`avg_over_time(sum(container_fs_limit_bytes{device=~"/dev/(nvme|sda).*", id="/", %s}) by (instance, device, %s)[%s:%dm])`, env.GetPromClusterFilter(), env.GetPromClusterLabel(), durStr, minsPerResolution)
		queryLocalActiveMins := fmt.Sprintf(`count(node_total_hourly_cost{%s}) by (%s, node)[%s:%dm]`, env.GetPromClusterFilter(), env.GetPromClusterLabel(), durStr, minsPerResolution)

		resChLocalStorageCost := ctx.QueryAtTime(queryLocalStorageCost, t)
		resChLocalStorageUsedCost := ctx.QueryAtTime(queryLocalStorageUsedCost, t)
		resChLocalStoreageUsedAvg := ctx.QueryAtTime(queryLocalStorageUsedAvg, t)
		resChLocalStoreageUsedMax := ctx.QueryAtTime(queryLocalStorageUsedMax, t)
		resChLocalStorageBytes := ctx.QueryAtTime(queryLocalStorageBytes, t)
		resChLocalActiveMins := ctx.QueryAtTime(queryLocalActiveMins, t)

		resLocalStorageCost, _ = resChLocalStorageCost.Await()
		resLocalStorageUsedCost, _ = resChLocalStorageUsedCost.Await()
		resLocalStorageUsedAvg, _ = resChLocalStoreageUsedAvg.Await()
		resLocalStorageUsedMax, _ = resChLocalStoreageUsedMax.Await()
		resLocalStorageBytes, _ = resChLocalStorageBytes.Await()
		resLocalActiveMins, _ = resChLocalActiveMins.Await()
	}

	if ctx.HasErrors() {
		return nil, ctx.ErrorCollection()
	}

	diskMap := map[DiskIdentifier]*Disk{}

	for _, result := range resPVCInfo {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		volumeName, err := result.GetString("volumename")
		if err != nil {
			log.Debugf("ClusterDisks: pv claim data missing volumename")
			continue
		}
		claimName, err := result.GetString("persistentvolumeclaim")
		if err != nil {
			log.Debugf("ClusterDisks: pv claim data missing persistentvolumeclaim")
			continue
		}
		claimNamespace, err := result.GetString("namespace")
		if err != nil {
			log.Debugf("ClusterDisks: pv claim data missing namespace")
			continue
		}

		key := DiskIdentifier{cluster, volumeName}
		if _, ok := diskMap[key]; !ok {
			diskMap[key] = &Disk{
				Cluster:   cluster,
				Name:      volumeName,
				Breakdown: &ClusterCostsBreakdown{},
			}
		}

		diskMap[key].VolumeName = volumeName
		diskMap[key].ClaimName = claimName
		diskMap[key].ClaimNamespace = claimNamespace
	}

	pvCosts(diskMap, resolution, resActiveMins, resPVSize, resPVCost, resPVUsedAvg, resPVUsedMax, resPVCInfo, cp, opencost.NewClosedWindow(start, end))

	type localStorage struct {
		device string
		disk   *Disk
	}

	localStorageDisks := map[DiskIdentifier]localStorage{}

	// Start with local storage bytes so that the device with the largest size which has passed the
	// query filters can be determined
	for _, result := range resLocalStorageBytes {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		name, err := result.GetString("instance")
		if err != nil {
			log.Warnf("ClusterDisks: local storage data missing instance")
			continue
		}

		device, err := result.GetString("device")
		if err != nil {
			log.Warnf("ClusterDisks: local storage data missing device")
			continue
		}

		bytes := result.Values[0].Value
		// Ignore disks that are larger than the max size
		if bytes > MAX_LOCAL_STORAGE_SIZE {
			continue
		}

		key := DiskIdentifier{cluster, name}

		// only keep the device with the most bytes per instance
		if current, ok := localStorageDisks[key]; !ok || current.disk.Bytes < bytes {
			localStorageDisks[key] = localStorage{
				device: device,
				disk: &Disk{
					Cluster:      cluster,
					Name:         name,
					Breakdown:    &ClusterCostsBreakdown{},
					Local:        true,
					StorageClass: opencost.LocalStorageClass,
					Bytes:        bytes,
				},
			}
		}
	}

	for _, result := range resLocalStorageCost {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		name, err := result.GetString("instance")
		if err != nil {
			log.Warnf("ClusterDisks: local storage data missing instance")
			continue
		}

		device, err := result.GetString("device")
		if err != nil {
			log.Warnf("ClusterDisks: local storage data missing device")
			continue
		}

		cost := result.Values[0].Value
		key := DiskIdentifier{cluster, name}
		ls, ok := localStorageDisks[key]
		if !ok || ls.device != device {
			continue
		}
		ls.disk.Cost = cost

	}

	for _, result := range resLocalStorageUsedCost {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		name, err := result.GetString("instance")
		if err != nil {
			log.Warnf("ClusterDisks: local storage usage data missing instance")
			continue
		}

		device, err := result.GetString("device")
		if err != nil {
			log.Warnf("ClusterDisks: local storage data missing device")
			continue
		}

		cost := result.Values[0].Value
		key := DiskIdentifier{cluster, name}
		ls, ok := localStorageDisks[key]
		if !ok || ls.device != device {
			continue
		}
		ls.disk.Breakdown.System = cost / ls.disk.Cost
	}

	for _, result := range resLocalStorageUsedAvg {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		name, err := result.GetString("instance")
		if err != nil {
			log.Warnf("ClusterDisks: local storage data missing instance")
			continue
		}

		device, err := result.GetString("device")
		if err != nil {
			log.Warnf("ClusterDisks: local storage data missing device")
			continue
		}

		bytesAvg := result.Values[0].Value
		key := DiskIdentifier{cluster, name}
		ls, ok := localStorageDisks[key]
		if !ok || ls.device != device {
			continue
		}
		ls.disk.BytesUsedAvgPtr = &bytesAvg
	}

	for _, result := range resLocalStorageUsedMax {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		name, err := result.GetString("instance")
		if err != nil {
			log.Warnf("ClusterDisks: local storage data missing instance")
			continue
		}

		device, err := result.GetString("device")
		if err != nil {
			log.Warnf("ClusterDisks: local storage data missing device")
			continue
		}

		bytesMax := result.Values[0].Value
		key := DiskIdentifier{cluster, name}
		ls, ok := localStorageDisks[key]
		if !ok || ls.device != device {
			continue
		}
		ls.disk.BytesUsedMaxPtr = &bytesMax
	}

	for _, result := range resLocalActiveMins {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		name, err := result.GetString("node")
		if err != nil {
			log.DedupedWarningf(5, "ClusterDisks: local active mins data missing instance")
			continue
		}

		providerID, err := result.GetString("provider_id")
		if err != nil {
			log.DedupedWarningf(5, "ClusterDisks: local active mins data missing instance")
			continue
		}

		key := DiskIdentifier{cluster, name}
		ls, ok := localStorageDisks[key]
		if !ok {
			continue
		}

		ls.disk.ProviderID = provider.ParseLocalDiskID(providerID)

		if len(result.Values) == 0 {
			continue
		}

		s := time.Unix(int64(result.Values[0].Timestamp), 0)
		e := time.Unix(int64(result.Values[len(result.Values)-1].Timestamp), 0)
		mins := e.Sub(s).Minutes()

		// TODO niko/assets if mins >= threshold, interpolate for missing data?

		ls.disk.End = e
		ls.disk.Start = s
		ls.disk.Minutes = mins
	}

	// move local storage disks to main disk map
	for key, ls := range localStorageDisks {
		diskMap[key] = ls.disk
	}

	var unTracedDiskLogData []DiskIdentifier
	//Iterating through Persistent Volume given by custom metrics kubecost_pv_info and assign the storage class if known and __unknown__ if not populated.
	for _, result := range resPVStorageClass {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		name, _ := result.GetString("persistentvolume")

		key := DiskIdentifier{cluster, name}
		if _, ok := diskMap[key]; !ok {
			if !slices.Contains(unTracedDiskLogData, key) {
				unTracedDiskLogData = append(unTracedDiskLogData, key)
			}
			continue
		}

		if len(result.Values) == 0 {
			continue
		}

		storageClass, err := result.GetString("storageclass")

		if err != nil {
			diskMap[key].StorageClass = opencost.UnknownStorageClass
		} else {
			diskMap[key].StorageClass = storageClass
		}
	}

	// Logging the unidentified disk information outside the loop

	for _, unIdentifiedDisk := range unTracedDiskLogData {
		log.Warnf("ClusterDisks: Cluster %s has Storage Class information for unidentified disk %s or disk deleted from analysis", unIdentifiedDisk.Cluster, unIdentifiedDisk.Name)
	}

	for _, disk := range diskMap {
		// Apply all remaining RAM to Idle
		disk.Breakdown.Idle = 1.0 - (disk.Breakdown.System + disk.Breakdown.Other + disk.Breakdown.User)

		// Set provider Id to the name for reconciliation
		if disk.ProviderID == "" {
			disk.ProviderID = disk.Name
		}
	}

	if !env.GetAssetIncludeLocalDiskCost() {
		return filterOutLocalPVs(diskMap), nil
	}

	return diskMap, nil
}

type NodeOverhead struct {
	CpuOverheadFraction float64
	RamOverheadFraction float64
}
type Node struct {
	Cluster         string
	Name            string
	ProviderID      string
	NodeType        string
	CPUCost         float64
	CPUCores        float64
	GPUCost         float64
	GPUCount        float64
	RAMCost         float64
	RAMBytes        float64
	Discount        float64
	Preemptible     bool
	CPUBreakdown    *ClusterCostsBreakdown
	RAMBreakdown    *ClusterCostsBreakdown
	Start           time.Time
	End             time.Time
	Minutes         float64
	Labels          map[string]string
	CostPerCPUHr    float64
	CostPerRAMGiBHr float64
	CostPerGPUHr    float64
	Overhead        *NodeOverhead
}

// GKE lies about the number of cores e2 nodes have. This table
// contains a mapping from node type -> actual CPU cores
// for those cases.
var partialCPUMap = map[string]float64{
	"e2-micro":  0.25,
	"e2-small":  0.5,
	"e2-medium": 1.0,
}

type NodeIdentifier struct {
	Cluster    string
	Name       string
	ProviderID string
}

type nodeIdentifierNoProviderID struct {
	Cluster string
	Name    string
}

func costTimesMinuteAndCount(activeDataMap map[NodeIdentifier]activeData, costMap map[NodeIdentifier]float64, resourceCountMap map[nodeIdentifierNoProviderID]float64) {
	for k, v := range activeDataMap {
		keyNon := nodeIdentifierNoProviderID{
			Cluster: k.Cluster,
			Name:    k.Name,
		}
		if cost, ok := costMap[k]; ok {
			minutes := v.minutes
			count := 1.0
			if c, ok := resourceCountMap[keyNon]; ok {
				count = c
			}
			costMap[k] = cost * (minutes / 60) * count
		}
	}
}

func costTimesMinute(activeDataMap map[NodeIdentifier]activeData, costMap map[NodeIdentifier]float64) {
	for k, v := range activeDataMap {
		if cost, ok := costMap[k]; ok {
			minutes := v.minutes
			costMap[k] = cost * (minutes / 60)
		}
	}
}

func ClusterNodes(cp models.Provider, client prometheus.Client, start, end time.Time) (map[NodeIdentifier]*Node, error) {
	// Start from the time "end", querying backwards
	t := end

	// minsPerResolution determines accuracy and resource use for the following
	// queries. Smaller values (higher resolution) result in better accuracy,
	// but more expensive queries, and vice-a-versa.
	resolution := env.GetETLResolution()
	//Ensuring if ETL_RESOLUTION_SECONDS is less than 60s default it to 1m
	var minsPerResolution int
	if minsPerResolution = int(resolution.Minutes()); int(resolution.Minutes()) == 0 {
		minsPerResolution = 1
		log.DedupedWarningf(3, "ClusterNodes(): Configured ETL resolution (%d seconds) is below the 60 seconds threshold. Overriding with 1 minute.", int(resolution.Seconds()))
	}

	durStr := timeutil.DurationString(end.Sub(start))
	if durStr == "" {
		return nil, fmt.Errorf("illegal duration value for %s", opencost.NewClosedWindow(start, end))
	}

	requiredCtx := prom.NewNamedContext(client, prom.ClusterContextName)
	optionalCtx := prom.NewNamedContext(client, prom.ClusterOptionalContextName)

	queryNodeCPUHourlyCost := fmt.Sprintf(`avg(avg_over_time(node_cpu_hourly_cost{%s}[%s])) by (%s, node, instance_type, provider_id)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryNodeCPUCoresCapacity := fmt.Sprintf(`avg(avg_over_time(kube_node_status_capacity_cpu_cores{%s}[%s])) by (%s, node)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryNodeCPUCoresAllocatable := fmt.Sprintf(`avg(avg_over_time(kube_node_status_allocatable_cpu_cores{%s}[%s])) by (%s, node)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryNodeRAMHourlyCost := fmt.Sprintf(`avg(avg_over_time(node_ram_hourly_cost{%s}[%s])) by (%s, node, instance_type, provider_id) / 1024 / 1024 / 1024`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryNodeRAMBytesCapacity := fmt.Sprintf(`avg(avg_over_time(kube_node_status_capacity_memory_bytes{%s}[%s])) by (%s, node)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryNodeRAMBytesAllocatable := fmt.Sprintf(`avg(avg_over_time(kube_node_status_allocatable_memory_bytes{%s}[%s])) by (%s, node)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryNodeGPUCount := fmt.Sprintf(`avg(avg_over_time(node_gpu_count{%s}[%s])) by (%s, node, provider_id)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryNodeGPUHourlyCost := fmt.Sprintf(`avg(avg_over_time(node_gpu_hourly_cost{%s}[%s])) by (%s, node, instance_type, provider_id)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryNodeCPUModeTotal := fmt.Sprintf(`sum(rate(node_cpu_seconds_total{%s}[%s:%dm])) by (kubernetes_node, %s, mode)`, env.GetPromClusterFilter(), durStr, minsPerResolution, env.GetPromClusterLabel())
	queryNodeRAMSystemPct := fmt.Sprintf(`sum(sum_over_time(container_memory_working_set_bytes{container_name!="POD",container_name!="",namespace="kube-system", %s}[%s:%dm])) by (instance, %s) / avg(label_replace(sum(sum_over_time(kube_node_status_capacity_memory_bytes{%s}[%s:%dm])) by (node, %s), "instance", "$1", "node", "(.*)")) by (instance, %s)`, env.GetPromClusterFilter(), durStr, minsPerResolution, env.GetPromClusterLabel(), env.GetPromClusterFilter(), durStr, minsPerResolution, env.GetPromClusterLabel(), env.GetPromClusterLabel())
	queryNodeRAMUserPct := fmt.Sprintf(`sum(sum_over_time(container_memory_working_set_bytes{container_name!="POD",container_name!="",namespace!="kube-system", %s}[%s:%dm])) by (instance, %s) / avg(label_replace(sum(sum_over_time(kube_node_status_capacity_memory_bytes{%s}[%s:%dm])) by (node, %s), "instance", "$1", "node", "(.*)")) by (instance, %s)`, env.GetPromClusterFilter(), durStr, minsPerResolution, env.GetPromClusterLabel(), env.GetPromClusterFilter(), durStr, minsPerResolution, env.GetPromClusterLabel(), env.GetPromClusterLabel())
	queryActiveMins := fmt.Sprintf(`avg(node_total_hourly_cost{%s}) by (node, %s, provider_id)[%s:%dm]`, env.GetPromClusterFilter(), env.GetPromClusterLabel(), durStr, minsPerResolution)
	queryIsSpot := fmt.Sprintf(`avg_over_time(kubecost_node_is_spot{%s}[%s:%dm])`, env.GetPromClusterFilter(), durStr, minsPerResolution)
	queryLabels := fmt.Sprintf(`count_over_time(kube_node_labels{%s}[%s:%dm])`, env.GetPromClusterFilter(), durStr, minsPerResolution)

	// Return errors if these fail
	resChNodeCPUHourlyCost := requiredCtx.QueryAtTime(queryNodeCPUHourlyCost, t)
	resChNodeCPUCoresCapacity := requiredCtx.QueryAtTime(queryNodeCPUCoresCapacity, t)
	resChNodeCPUCoresAllocatable := requiredCtx.QueryAtTime(queryNodeCPUCoresAllocatable, t)
	resChNodeRAMHourlyCost := requiredCtx.QueryAtTime(queryNodeRAMHourlyCost, t)
	resChNodeRAMBytesCapacity := requiredCtx.QueryAtTime(queryNodeRAMBytesCapacity, t)
	resChNodeRAMBytesAllocatable := requiredCtx.QueryAtTime(queryNodeRAMBytesAllocatable, t)
	resChNodeGPUCount := requiredCtx.QueryAtTime(queryNodeGPUCount, t)
	resChNodeGPUHourlyCost := requiredCtx.QueryAtTime(queryNodeGPUHourlyCost, t)
	resChActiveMins := requiredCtx.QueryAtTime(queryActiveMins, t)
	resChIsSpot := requiredCtx.QueryAtTime(queryIsSpot, t)

	// Do not return errors if these fail, but log warnings
	resChNodeCPUModeTotal := optionalCtx.QueryAtTime(queryNodeCPUModeTotal, t)
	resChNodeRAMSystemPct := optionalCtx.QueryAtTime(queryNodeRAMSystemPct, t)
	resChNodeRAMUserPct := optionalCtx.QueryAtTime(queryNodeRAMUserPct, t)
	resChLabels := optionalCtx.QueryAtTime(queryLabels, t)

	resNodeCPUHourlyCost, _ := resChNodeCPUHourlyCost.Await()
	resNodeCPUCoresCapacity, _ := resChNodeCPUCoresCapacity.Await()
	resNodeCPUCoresAllocatable, _ := resChNodeCPUCoresAllocatable.Await()
	resNodeGPUCount, _ := resChNodeGPUCount.Await()
	resNodeGPUHourlyCost, _ := resChNodeGPUHourlyCost.Await()
	resNodeRAMHourlyCost, _ := resChNodeRAMHourlyCost.Await()
	resNodeRAMBytesCapacity, _ := resChNodeRAMBytesCapacity.Await()
	resNodeRAMBytesAllocatable, _ := resChNodeRAMBytesAllocatable.Await()
	resIsSpot, _ := resChIsSpot.Await()
	resNodeCPUModeTotal, _ := resChNodeCPUModeTotal.Await()
	resNodeRAMSystemPct, _ := resChNodeRAMSystemPct.Await()
	resNodeRAMUserPct, _ := resChNodeRAMUserPct.Await()
	resActiveMins, _ := resChActiveMins.Await()
	resLabels, _ := resChLabels.Await()

	if optionalCtx.HasErrors() {
		for _, err := range optionalCtx.Errors() {
			log.Warnf("ClusterNodes: %s", err)
		}
	}
	if requiredCtx.HasErrors() {
		for _, err := range requiredCtx.Errors() {
			log.Errorf("ClusterNodes: %s", err)
		}

		return nil, requiredCtx.ErrorCollection()
	}

	activeDataMap := buildActiveDataMap(resActiveMins, resolution, opencost.NewClosedWindow(start, end))

	gpuCountMap := buildGPUCountMap(resNodeGPUCount)
	preemptibleMap := buildPreemptibleMap(resIsSpot)

	cpuCostMap, clusterAndNameToType1 := buildCPUCostMap(resNodeCPUHourlyCost, cp, preemptibleMap)
	ramCostMap, clusterAndNameToType2 := buildRAMCostMap(resNodeRAMHourlyCost, cp, preemptibleMap)
	gpuCostMap, clusterAndNameToType3 := buildGPUCostMap(resNodeGPUHourlyCost, gpuCountMap, cp, preemptibleMap)

	clusterAndNameToTypeIntermediate := mergeTypeMaps(clusterAndNameToType1, clusterAndNameToType2)
	clusterAndNameToType := mergeTypeMaps(clusterAndNameToTypeIntermediate, clusterAndNameToType3)

	cpuCoresCapacityMap := buildCPUCoresMap(resNodeCPUCoresCapacity)
	ramBytesCapacityMap := buildRAMBytesMap(resNodeRAMBytesCapacity)

	cpuCoresAllocatableMap := buildCPUCoresMap(resNodeCPUCoresAllocatable)
	ramBytesAllocatableMap := buildRAMBytesMap(resNodeRAMBytesAllocatable)
	overheadMap := buildOverheadMap(ramBytesCapacityMap, ramBytesAllocatableMap, cpuCoresCapacityMap, cpuCoresAllocatableMap)

	ramUserPctMap := buildRAMUserPctMap(resNodeRAMUserPct)
	ramSystemPctMap := buildRAMSystemPctMap(resNodeRAMSystemPct)

	cpuBreakdownMap := buildCPUBreakdownMap(resNodeCPUModeTotal)

	labelsMap := buildLabelsMap(resLabels)

	costTimesMinuteAndCount(activeDataMap, cpuCostMap, cpuCoresCapacityMap)
	costTimesMinuteAndCount(activeDataMap, ramCostMap, ramBytesCapacityMap)
	costTimesMinute(activeDataMap, gpuCostMap) // there's no need to do a weird "nodeIdentifierNoProviderID" type match since gpuCounts have a providerID

	nodeMap := buildNodeMap(
		cpuCostMap, ramCostMap, gpuCostMap, gpuCountMap,
		cpuCoresCapacityMap, ramBytesCapacityMap, ramUserPctMap,
		ramSystemPctMap,
		cpuBreakdownMap,
		activeDataMap,
		preemptibleMap,
		labelsMap,
		clusterAndNameToType,
		resolution,
		overheadMap,
	)

	c, err := cp.GetConfig()
	if err != nil {
		return nil, err
	}

	discount, err := ParsePercentString(c.Discount)
	if err != nil {
		return nil, err
	}

	negotiatedDiscount, err := ParsePercentString(c.NegotiatedDiscount)
	if err != nil {
		return nil, err
	}

	for _, node := range nodeMap {
		// TODO take GKE Reserved Instances into account
		node.Discount = cp.CombinedDiscountForNode(node.NodeType, node.Preemptible, discount, negotiatedDiscount)

		// Apply all remaining resources to Idle
		node.CPUBreakdown.Idle = 1.0 - (node.CPUBreakdown.System + node.CPUBreakdown.Other + node.CPUBreakdown.User)
		node.RAMBreakdown.Idle = 1.0 - (node.RAMBreakdown.System + node.RAMBreakdown.Other + node.RAMBreakdown.User)
	}

	return nodeMap, nil
}

type LoadBalancerIdentifier struct {
	Cluster   string
	Namespace string
	Name      string
}

type LoadBalancer struct {
	Cluster    string
	Namespace  string
	Name       string
	ProviderID string
	Cost       float64
	Start      time.Time
	End        time.Time
	Minutes    float64
	Private    bool
	Ip         string
}

func ClusterLoadBalancers(client prometheus.Client, start, end time.Time) (map[LoadBalancerIdentifier]*LoadBalancer, error) {

	// Start from the time "end", querying backwards
	t := end

	// minsPerResolution determines accuracy and resource use for the following
	// queries. Smaller values (higher resolution) result in better accuracy,
	// but more expensive queries, and vice-a-versa.
	resolution := env.GetETLResolution()
	//Ensuring if ETL_RESOLUTION_SECONDS is less than 60s default it to 1m
	var minsPerResolution int
	if minsPerResolution = int(resolution.Minutes()); int(resolution.Minutes()) == 0 {
		minsPerResolution = 1
		log.DedupedWarningf(3, "ClusterLoadBalancers(): Configured ETL resolution (%d seconds) is below the 60 seconds threshold. Overriding with 1 minute.", int(resolution.Seconds()))
	}

	// Query for the duration between start and end
	durStr := timeutil.DurationString(end.Sub(start))
	if durStr == "" {
		return nil, fmt.Errorf("illegal duration value for %s", opencost.NewClosedWindow(start, end))
	}

	ctx := prom.NewNamedContext(client, prom.ClusterContextName)

	queryLBCost := fmt.Sprintf(`avg(avg_over_time(kubecost_load_balancer_cost{%s}[%s])) by (namespace, service_name, %s, ingress_ip)`, env.GetPromClusterFilter(), durStr, env.GetPromClusterLabel())
	queryActiveMins := fmt.Sprintf(`avg(kubecost_load_balancer_cost{%s}) by (namespace, service_name, %s, ingress_ip)[%s:%dm]`, env.GetPromClusterFilter(), env.GetPromClusterLabel(), durStr, minsPerResolution)

	resChLBCost := ctx.QueryAtTime(queryLBCost, t)
	resChActiveMins := ctx.QueryAtTime(queryActiveMins, t)

	resLBCost, _ := resChLBCost.Await()
	resActiveMins, _ := resChActiveMins.Await()

	if ctx.HasErrors() {
		return nil, ctx.ErrorCollection()
	}

	loadBalancerMap := make(map[LoadBalancerIdentifier]*LoadBalancer, len(resActiveMins))

	for _, result := range resActiveMins {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}
		namespace, err := result.GetString("namespace")
		if err != nil {
			log.Warnf("ClusterLoadBalancers: LB cost data missing namespace")
			continue
		}
		name, err := result.GetString("service_name")
		if err != nil {
			log.Warnf("ClusterLoadBalancers: LB cost data missing service_name")
			continue
		}
		providerID, err := result.GetString("ingress_ip")
		if err != nil {
			log.DedupedWarningf(5, "ClusterLoadBalancers: LB cost data missing ingress_ip")
			providerID = ""
		}

		key := LoadBalancerIdentifier{
			Cluster:   cluster,
			Namespace: namespace,
			Name:      name,
		}

		// Skip if there are no data
		if len(result.Values) == 0 {
			continue
		}

		// Add load balancer to the set of load balancers
		if _, ok := loadBalancerMap[key]; !ok {
			loadBalancerMap[key] = &LoadBalancer{
				Cluster:    cluster,
				Namespace:  namespace,
				Name:       fmt.Sprintf("%s/%s", namespace, name), // TODO:ETL this is kept for backwards-compatibility, but not good
				ProviderID: provider.ParseLBID(providerID),
			}
		}

		// Append start, end, and minutes. This should come before all other data.
		s := time.Unix(int64(result.Values[0].Timestamp), 0)
		e := time.Unix(int64(result.Values[len(result.Values)-1].Timestamp), 0)
		loadBalancerMap[key].Start = s
		loadBalancerMap[key].End = e
		loadBalancerMap[key].Minutes = e.Sub(s).Minutes()

		// Fill in Provider ID if it is available and missing in the loadBalancerMap
		// Prevents there from being a duplicate LoadBalancers on the same day
		if providerID != "" && loadBalancerMap[key].ProviderID == "" {
			loadBalancerMap[key].ProviderID = providerID
		}
	}

	for _, result := range resLBCost {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}
		namespace, err := result.GetString("namespace")
		if err != nil {
			log.Warnf("ClusterLoadBalancers: LB cost data missing namespace")
			continue
		}
		name, err := result.GetString("service_name")
		if err != nil {
			log.Warnf("ClusterLoadBalancers: LB cost data missing service_name")
			continue
		}

		providerID, err := result.GetString("ingress_ip")
		if err != nil {
			log.DedupedWarningf(5, "ClusterLoadBalancers: LB cost data missing ingress_ip")
			// only update asset cost when an actual IP was returned
			continue
		}
		key := LoadBalancerIdentifier{
			Cluster:   cluster,
			Namespace: namespace,
			Name:      name,
		}

		// Apply cost as price-per-hour * hours
		if lb, ok := loadBalancerMap[key]; ok {
			lbPricePerHr := result.Values[0].Value

			// interpolate any missing data
			resultMins := lb.Minutes
			if resultMins > 0 {
				scaleFactor := (resultMins + resolution.Minutes()) / resultMins

				hrs := (lb.Minutes * scaleFactor) / 60.0
				lb.Cost += lbPricePerHr * hrs
			} else {
				log.DedupedWarningf(20, "ClusterLoadBalancers: found zero minutes for key: %v", key)
			}

			if lb.Ip != "" && lb.Ip != providerID {
				log.DedupedWarningf(5, "ClusterLoadBalancers: multiple IPs per load balancer not supported, using most recent IP")
			}
			lb.Ip = providerID

			lb.Private = privateIPCheck(providerID)
		} else {
			log.DedupedWarningf(20, "ClusterLoadBalancers: found minutes for key that does not exist: %v", key)
		}
	}

	return loadBalancerMap, nil
}

// Check if an ip is private.
func privateIPCheck(ip string) bool {
	ipAddress := net.ParseIP(ip)
	return ipAddress.IsPrivate()
}

// ComputeClusterCosts gives the cumulative and monthly-rate cluster costs over a window of time for all clusters.
func (a *Accesses) ComputeClusterCosts(client prometheus.Client, provider models.Provider, window, offset time.Duration, withBreakdown bool) (map[string]*ClusterCosts, error) {
	if window < 10*time.Minute {
		return nil, fmt.Errorf("minimum window of 10m required; got %s", window)
	}

	// Compute number of minutes in the full interval, for use interpolating missed scrapes or scaling missing data
	start, end := timeutil.ParseTimeRange(window, offset)

	mins := end.Sub(start).Minutes()

	// minsPerResolution determines accuracy and resource use for the following
	// queries. Smaller values (higher resolution) result in better accuracy,
	// but more expensive queries, and vice-a-versa.
	resolution := env.GetETLResolution()
	//Ensuring if ETL_RESOLUTION_SECONDS is less than 60s default it to 1m
	var minsPerResolution int
	if minsPerResolution = int(resolution.Minutes()); int(resolution.Minutes()) < 1 {
		minsPerResolution = 1
		log.DedupedWarningf(3, "ComputeClusterCosts(): Configured ETL resolution (%d seconds) is below the 60 seconds threshold. Overriding with 1 minute.", int(resolution.Seconds()))
	}

	windowStr := timeutil.DurationString(window)

	// hourlyToCumulative is a scaling factor that, when multiplied by an hourly
	// value, converts it to a cumulative value; i.e.
	// [$/hr] * [min/res]*[hr/min] = [$/res]
	hourlyToCumulative := float64(minsPerResolution) * (1.0 / 60.0)

	const fmtQueryDataCount = `
		count_over_time(sum(kube_node_status_capacity_cpu_cores{%s}) by (%s)[%s:%dm]%s) * %d
	`

	const fmtQueryTotalGPU = `
		sum(
			sum_over_time(node_gpu_hourly_cost{%s}[%s:%dm]%s) * %f
		) by (%s)
	`

	const fmtQueryTotalCPU = `
		sum(
			sum_over_time(avg(kube_node_status_capacity_cpu_cores{%s}) by (node, %s)[%s:%dm]%s) *
			avg(avg_over_time(node_cpu_hourly_cost{%s}[%s:%dm]%s)) by (node, %s) * %f
		) by (%s)
	`

	const fmtQueryTotalRAM = `
		sum(
			sum_over_time(avg(kube_node_status_capacity_memory_bytes{%s}) by (node, %s)[%s:%dm]%s) / 1024 / 1024 / 1024 *
			avg(avg_over_time(node_ram_hourly_cost{%s}[%s:%dm]%s)) by (node, %s) * %f
		) by (%s)
	`

	const fmtQueryTotalStorage = `
		sum(
			sum_over_time(avg(kube_persistentvolume_capacity_bytes{%s}) by (persistentvolume, %s)[%s:%dm]%s) / 1024 / 1024 / 1024 *
			avg(avg_over_time(pv_hourly_cost{%s}[%s:%dm]%s)) by (persistentvolume, %s) * %f
		) by (%s)
	`

	const fmtQueryCPUModePct = `
		sum(rate(node_cpu_seconds_total{%s}[%s]%s)) by (%s, mode) / ignoring(mode)
		group_left sum(rate(node_cpu_seconds_total{%s}[%s]%s)) by (%s)
	`

	const fmtQueryRAMSystemPct = `
		sum(sum_over_time(container_memory_usage_bytes{container_name!="",namespace="kube-system", %s}[%s:%dm]%s)) by (%s)
		/ sum(sum_over_time(kube_node_status_capacity_memory_bytes{%s}[%s:%dm]%s)) by (%s)
	`

	const fmtQueryRAMUserPct = `
		sum(sum_over_time(kubecost_cluster_memory_working_set_bytes{%s}[%s:%dm]%s)) by (%s)
		/ sum(sum_over_time(kube_node_status_capacity_memory_bytes{%s}[%s:%dm]%s)) by (%s)
	`

	// TODO niko/clustercost metric "kubelet_volume_stats_used_bytes" was deprecated in 1.12, then seems to have come back in 1.17
	// const fmtQueryPVStorageUsePct = `(sum(kube_persistentvolumeclaim_info) by (persistentvolumeclaim, storageclass,namespace) + on (persistentvolumeclaim,namespace)
	// group_right(storageclass) sum(kubelet_volume_stats_used_bytes) by (persistentvolumeclaim,namespace))`

	queryUsedLocalStorage := provider.GetLocalStorageQuery(window, offset, false, true)

	queryTotalLocalStorage := provider.GetLocalStorageQuery(window, offset, false, false)
	if queryTotalLocalStorage != "" {
		queryTotalLocalStorage = fmt.Sprintf(" + %s", queryTotalLocalStorage)
	}

	fmtOffset := timeutil.DurationToPromOffsetString(offset)

	queryDataCount := fmt.Sprintf(fmtQueryDataCount, env.GetPromClusterFilter(), env.GetPromClusterLabel(), windowStr, minsPerResolution, fmtOffset, minsPerResolution)
	queryTotalGPU := fmt.Sprintf(fmtQueryTotalGPU, env.GetPromClusterFilter(), windowStr, minsPerResolution, fmtOffset, hourlyToCumulative, env.GetPromClusterLabel())
	queryTotalCPU := fmt.Sprintf(fmtQueryTotalCPU, env.GetPromClusterFilter(), env.GetPromClusterLabel(), windowStr, minsPerResolution, fmtOffset, env.GetPromClusterFilter(), windowStr, minsPerResolution, fmtOffset, env.GetPromClusterLabel(), hourlyToCumulative, env.GetPromClusterLabel())
	queryTotalRAM := fmt.Sprintf(fmtQueryTotalRAM, env.GetPromClusterFilter(), env.GetPromClusterLabel(), windowStr, minsPerResolution, fmtOffset, env.GetPromClusterFilter(), windowStr, minsPerResolution, fmtOffset, env.GetPromClusterLabel(), hourlyToCumulative, env.GetPromClusterLabel())
	queryTotalStorage := fmt.Sprintf(fmtQueryTotalStorage, env.GetPromClusterFilter(), env.GetPromClusterLabel(), windowStr, minsPerResolution, fmtOffset, env.GetPromClusterFilter(), windowStr, minsPerResolution, fmtOffset, env.GetPromClusterLabel(), hourlyToCumulative, env.GetPromClusterLabel())

	ctx := prom.NewNamedContext(client, prom.ClusterContextName)

	resChs := ctx.QueryAll(
		queryDataCount,
		queryTotalGPU,
		queryTotalCPU,
		queryTotalRAM,
		queryTotalStorage,
	)

	// Only submit the local storage query if it is valid. Otherwise Prometheus
	// will return errors. Always append something to resChs, regardless, to
	// maintain indexing.
	if queryTotalLocalStorage != "" {
		resChs = append(resChs, ctx.Query(queryTotalLocalStorage))
	} else {
		resChs = append(resChs, nil)
	}

	if withBreakdown {
		queryCPUModePct := fmt.Sprintf(fmtQueryCPUModePct, env.GetPromClusterFilter(), windowStr, fmtOffset, env.GetPromClusterLabel(), env.GetPromClusterFilter(), windowStr, fmtOffset, env.GetPromClusterLabel())
		queryRAMSystemPct := fmt.Sprintf(fmtQueryRAMSystemPct, env.GetPromClusterFilter(), windowStr, minsPerResolution, fmtOffset, env.GetPromClusterLabel(), env.GetPromClusterFilter(), windowStr, minsPerResolution, fmtOffset, env.GetPromClusterLabel())
		queryRAMUserPct := fmt.Sprintf(fmtQueryRAMUserPct, env.GetPromClusterFilter(), windowStr, minsPerResolution, fmtOffset, env.GetPromClusterLabel(), env.GetPromClusterFilter(), windowStr, minsPerResolution, fmtOffset, env.GetPromClusterLabel())

		bdResChs := ctx.QueryAll(
			queryCPUModePct,
			queryRAMSystemPct,
			queryRAMUserPct,
		)

		// Only submit the local storage query if it is valid. Otherwise Prometheus
		// will return errors. Always append something to resChs, regardless, to
		// maintain indexing.
		if queryUsedLocalStorage != "" {
			bdResChs = append(bdResChs, ctx.Query(queryUsedLocalStorage))
		} else {
			bdResChs = append(bdResChs, nil)
		}

		resChs = append(resChs, bdResChs...)
	}

	resDataCount, _ := resChs[0].Await()
	resTotalGPU, _ := resChs[1].Await()
	resTotalCPU, _ := resChs[2].Await()
	resTotalRAM, _ := resChs[3].Await()
	resTotalStorage, _ := resChs[4].Await()
	if ctx.HasErrors() {
		return nil, ctx.ErrorCollection()
	}

	defaultClusterID := env.GetClusterID()

	dataMinsByCluster := map[string]float64{}
	for _, result := range resDataCount {
		clusterID, _ := result.GetString(env.GetPromClusterLabel())
		if clusterID == "" {
			clusterID = defaultClusterID
		}
		dataMins := mins
		if len(result.Values) > 0 {
			dataMins = result.Values[0].Value
		} else {
			log.Warnf("Cluster cost data count returned no results for cluster %s", clusterID)
		}
		dataMinsByCluster[clusterID] = dataMins
	}

	// Determine combined discount
	discount, customDiscount := 0.0, 0.0
	c, err := a.CloudProvider.GetConfig()
	if err == nil {
		discount, err = ParsePercentString(c.Discount)
		if err != nil {
			discount = 0.0
		}
		customDiscount, err = ParsePercentString(c.NegotiatedDiscount)
		if err != nil {
			customDiscount = 0.0
		}
	}

	// Intermediate structure storing mapping of [clusterID][type ∈ {cpu, ram, storage, total}]=cost
	costData := make(map[string]map[string]float64)

	// Helper function to iterate over Prom query results, parsing the raw values into
	// the intermediate costData structure.
	setCostsFromResults := func(costData map[string]map[string]float64, results []*prom.QueryResult, name string, discount float64, customDiscount float64) {
		for _, result := range results {
			clusterID, _ := result.GetString(env.GetPromClusterLabel())
			if clusterID == "" {
				clusterID = defaultClusterID
			}
			if _, ok := costData[clusterID]; !ok {
				costData[clusterID] = map[string]float64{}
			}
			if len(result.Values) > 0 {
				costData[clusterID][name] += result.Values[0].Value * (1.0 - discount) * (1.0 - customDiscount)
				costData[clusterID]["total"] += result.Values[0].Value * (1.0 - discount) * (1.0 - customDiscount)
			}
		}
	}
	// Apply both sustained use and custom discounts to RAM and CPU
	setCostsFromResults(costData, resTotalCPU, "cpu", discount, customDiscount)
	setCostsFromResults(costData, resTotalRAM, "ram", discount, customDiscount)
	// Apply only custom discount to GPU and storage
	setCostsFromResults(costData, resTotalGPU, "gpu", 0.0, customDiscount)
	setCostsFromResults(costData, resTotalStorage, "storage", 0.0, customDiscount)
	if queryTotalLocalStorage != "" {
		resTotalLocalStorage, err := resChs[5].Await()
		if err != nil {
			return nil, err
		}
		setCostsFromResults(costData, resTotalLocalStorage, "localstorage", 0.0, customDiscount)
	}

	cpuBreakdownMap := map[string]*ClusterCostsBreakdown{}
	ramBreakdownMap := map[string]*ClusterCostsBreakdown{}
	pvUsedCostMap := map[string]float64{}
	if withBreakdown {
		resCPUModePct, _ := resChs[6].Await()
		resRAMSystemPct, _ := resChs[7].Await()
		resRAMUserPct, _ := resChs[8].Await()
		if ctx.HasErrors() {
			return nil, ctx.ErrorCollection()
		}

		for _, result := range resCPUModePct {
			clusterID, _ := result.GetString(env.GetPromClusterLabel())
			if clusterID == "" {
				clusterID = defaultClusterID
			}
			if _, ok := cpuBreakdownMap[clusterID]; !ok {
				cpuBreakdownMap[clusterID] = &ClusterCostsBreakdown{}
			}
			cpuBD := cpuBreakdownMap[clusterID]

			mode, err := result.GetString("mode")
			if err != nil {
				log.Warnf("ComputeClusterCosts: unable to read CPU mode: %s", err)
				mode = "other"
			}

			switch mode {
			case "idle":
				cpuBD.Idle += result.Values[0].Value
			case "system":
				cpuBD.System += result.Values[0].Value
			case "user":
				cpuBD.User += result.Values[0].Value
			default:
				cpuBD.Other += result.Values[0].Value
			}
		}

		for _, result := range resRAMSystemPct {
			clusterID, _ := result.GetString(env.GetPromClusterLabel())
			if clusterID == "" {
				clusterID = defaultClusterID
			}
			if _, ok := ramBreakdownMap[clusterID]; !ok {
				ramBreakdownMap[clusterID] = &ClusterCostsBreakdown{}
			}
			ramBD := ramBreakdownMap[clusterID]
			ramBD.System += result.Values[0].Value
		}
		for _, result := range resRAMUserPct {
			clusterID, _ := result.GetString(env.GetPromClusterLabel())
			if clusterID == "" {
				clusterID = defaultClusterID
			}
			if _, ok := ramBreakdownMap[clusterID]; !ok {
				ramBreakdownMap[clusterID] = &ClusterCostsBreakdown{}
			}
			ramBD := ramBreakdownMap[clusterID]
			ramBD.User += result.Values[0].Value
		}
		for _, ramBD := range ramBreakdownMap {
			remaining := 1.0
			remaining -= ramBD.Other
			remaining -= ramBD.System
			remaining -= ramBD.User
			ramBD.Idle = remaining
		}

		if queryUsedLocalStorage != "" {
			resUsedLocalStorage, err := resChs[9].Await()
			if err != nil {
				return nil, err
			}
			for _, result := range resUsedLocalStorage {
				clusterID, _ := result.GetString(env.GetPromClusterLabel())
				if clusterID == "" {
					clusterID = defaultClusterID
				}
				pvUsedCostMap[clusterID] += result.Values[0].Value
			}
		}
	}

	if ctx.HasErrors() {
		for _, err := range ctx.Errors() {
			log.Errorf("ComputeClusterCosts: %s", err)
		}
		return nil, ctx.ErrorCollection()
	}

	// Convert intermediate structure to Costs instances
	costsByCluster := map[string]*ClusterCosts{}
	for id, cd := range costData {
		dataMins, ok := dataMinsByCluster[id]
		if !ok {
			dataMins = mins
			log.Warnf("Cluster cost data count not found for cluster %s", id)
		}
		costs, err := NewClusterCostsFromCumulative(cd["cpu"], cd["gpu"], cd["ram"], cd["storage"]+cd["localstorage"], window, offset, dataMins/timeutil.MinsPerHour)
		if err != nil {
			log.Warnf("Failed to parse cluster costs on %s (%s) from cumulative data: %+v", window, offset, cd)
			return nil, err
		}

		if cpuBD, ok := cpuBreakdownMap[id]; ok {
			costs.CPUBreakdown = cpuBD
		}
		if ramBD, ok := ramBreakdownMap[id]; ok {
			costs.RAMBreakdown = ramBD
		}
		costs.StorageBreakdown = &ClusterCostsBreakdown{}
		if pvUC, ok := pvUsedCostMap[id]; ok {
			costs.StorageBreakdown.Idle = (costs.StorageCumulative - pvUC) / costs.StorageCumulative
			costs.StorageBreakdown.User = pvUC / costs.StorageCumulative
		}
		costs.DataMinutes = dataMins
		costsByCluster[id] = costs
	}

	return costsByCluster, nil
}

type Totals struct {
	TotalCost   [][]string `json:"totalcost"`
	CPUCost     [][]string `json:"cpucost"`
	MemCost     [][]string `json:"memcost"`
	StorageCost [][]string `json:"storageCost"`
}

func resultToTotals(qrs []*prom.QueryResult) ([][]string, error) {
	if len(qrs) == 0 {
		return [][]string{}, fmt.Errorf("Not enough data available in the selected time range")
	}

	result := qrs[0]
	totals := [][]string{}
	for _, value := range result.Values {
		d0 := fmt.Sprintf("%f", value.Timestamp)
		d1 := fmt.Sprintf("%f", value.Value)
		toAppend := []string{
			d0,
			d1,
		}
		totals = append(totals, toAppend)
	}
	return totals, nil
}

// ClusterCostsOverTime gives the full cluster costs over time
func ClusterCostsOverTime(cli prometheus.Client, provider models.Provider, startString, endString string, window, offset time.Duration) (*Totals, error) {
	localStorageQuery := provider.GetLocalStorageQuery(window, offset, true, false)
	if localStorageQuery != "" {
		localStorageQuery = fmt.Sprintf("+ %s", localStorageQuery)
	}

	layout := "2006-01-02T15:04:05.000Z"

	start, err := time.Parse(layout, startString)
	if err != nil {
		log.Errorf("Error parsing time %s. Error: %s", startString, err.Error())
		return nil, err
	}
	end, err := time.Parse(layout, endString)
	if err != nil {
		log.Errorf("Error parsing time %s. Error: %s", endString, err.Error())
		return nil, err
	}
	fmtWindow := timeutil.DurationString(window)

	if fmtWindow == "" {
		err := fmt.Errorf("window value invalid or missing")
		log.Errorf("Error parsing time %v. Error: %s", window, err.Error())
		return nil, err
	}

	fmtOffset := timeutil.DurationToPromOffsetString(offset)

	qCores := fmt.Sprintf(queryClusterCores, env.GetPromClusterFilter(), fmtWindow, fmtOffset, env.GetPromClusterLabel(), env.GetPromClusterFilter(), fmtWindow, fmtOffset, env.GetPromClusterLabel(), env.GetPromClusterFilter(), fmtWindow, fmtOffset, env.GetPromClusterLabel(), env.GetPromClusterLabel())
	qRAM := fmt.Sprintf(queryClusterRAM, env.GetPromClusterFilter(), fmtWindow, fmtOffset, env.GetPromClusterLabel(), env.GetPromClusterFilter(), fmtWindow, fmtOffset, env.GetPromClusterLabel(), env.GetPromClusterLabel())
	qStorage := fmt.Sprintf(queryStorage, env.GetPromClusterFilter(), fmtWindow, fmtOffset, env.GetPromClusterLabel(), env.GetPromClusterFilter(), fmtWindow, fmtOffset, env.GetPromClusterLabel(), env.GetPromClusterLabel(), localStorageQuery)
	qTotal := fmt.Sprintf(queryTotal, env.GetPromClusterFilter(), env.GetPromClusterLabel(), env.GetPromClusterFilter(), env.GetPromClusterLabel(), env.GetPromClusterFilter(), env.GetPromClusterLabel(), env.GetPromClusterLabel(), localStorageQuery)

	ctx := prom.NewNamedContext(cli, prom.ClusterContextName)
	resChClusterCores := ctx.QueryRange(qCores, start, end, window)
	resChClusterRAM := ctx.QueryRange(qRAM, start, end, window)
	resChStorage := ctx.QueryRange(qStorage, start, end, window)
	resChTotal := ctx.QueryRange(qTotal, start, end, window)

	resultClusterCores, err := resChClusterCores.Await()
	if err != nil {
		return nil, err
	}

	resultClusterRAM, err := resChClusterRAM.Await()
	if err != nil {
		return nil, err
	}

	resultStorage, err := resChStorage.Await()
	if err != nil {
		return nil, err
	}

	resultTotal, err := resChTotal.Await()
	if err != nil {
		return nil, err
	}

	coreTotal, err := resultToTotals(resultClusterCores)
	if err != nil {
		log.Infof("[Warning] ClusterCostsOverTime: no cpu data: %s", err)
		return nil, err
	}

	ramTotal, err := resultToTotals(resultClusterRAM)
	if err != nil {
		log.Infof("[Warning] ClusterCostsOverTime: no ram data: %s", err)
		return nil, err
	}

	storageTotal, err := resultToTotals(resultStorage)
	if err != nil {
		log.Infof("[Warning] ClusterCostsOverTime: no storage data: %s", err)
	}

	clusterTotal, err := resultToTotals(resultTotal)
	if err != nil {
		// If clusterTotal query failed, it's likely because there are no PVs, which
		// causes the qTotal query to return no data. Instead, query only node costs.
		// If that fails, return an error because something is actually wrong.
		qNodes := fmt.Sprintf(queryNodes, env.GetPromClusterFilter(), env.GetPromClusterLabel(), localStorageQuery)

		resultNodes, warnings, err := ctx.QueryRangeSync(qNodes, start, end, window)
		for _, warning := range warnings {
			log.Warnf(warning)
		}
		if err != nil {
			return nil, err
		}

		clusterTotal, err = resultToTotals(resultNodes)
		if err != nil {
			log.Infof("[Warning] ClusterCostsOverTime: no node data: %s", err)
			return nil, err
		}
	}

	return &Totals{
		TotalCost:   clusterTotal,
		CPUCost:     coreTotal,
		MemCost:     ramTotal,
		StorageCost: storageTotal,
	}, nil
}

func pvCosts(diskMap map[DiskIdentifier]*Disk, resolution time.Duration, resActiveMins, resPVSize, resPVCost, resPVUsedAvg, resPVUsedMax, resPVCInfo []*prom.QueryResult, cp models.Provider, window opencost.Window) {
	for _, result := range resActiveMins {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		name, err := result.GetString("persistentvolume")
		if err != nil {
			log.Warnf("ClusterDisks: active mins missing pv name")
			continue
		}

		if len(result.Values) == 0 {
			continue
		}

		key := DiskIdentifier{cluster, name}
		if _, ok := diskMap[key]; !ok {
			diskMap[key] = &Disk{
				Cluster:   cluster,
				Name:      name,
				Breakdown: &ClusterCostsBreakdown{},
			}
		}

		s, e := calculateStartAndEnd(result, resolution, window)
		mins := e.Sub(s).Minutes()

		diskMap[key].End = e
		diskMap[key].Start = s
		diskMap[key].Minutes = mins
	}

	for _, result := range resPVSize {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		name, err := result.GetString("persistentvolume")
		if err != nil {
			log.Warnf("ClusterDisks: PV size data missing persistentvolume")
			continue
		}

		// TODO niko/assets storage class

		bytes := result.Values[0].Value
		key := DiskIdentifier{cluster, name}
		if _, ok := diskMap[key]; !ok {
			diskMap[key] = &Disk{
				Cluster:   cluster,
				Name:      name,
				Breakdown: &ClusterCostsBreakdown{},
			}
		}
		diskMap[key].Bytes = bytes
	}

	customPricingEnabled := provider.CustomPricesEnabled(cp)
	customPricingConfig, err := cp.GetConfig()
	if err != nil {
		log.Warnf("ClusterDisks: failed to load custom pricing: %s", err)
	}

	for _, result := range resPVCost {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		name, err := result.GetString("persistentvolume")
		if err != nil {
			log.Warnf("ClusterDisks: PV cost data missing persistentvolume")
			continue
		}

		// TODO niko/assets storage class

		var cost float64
		if customPricingEnabled && customPricingConfig != nil {
			customPVCostStr := customPricingConfig.Storage

			customPVCost, err := strconv.ParseFloat(customPVCostStr, 64)
			if err != nil {
				log.Warnf("ClusterDisks: error parsing custom PV price: %s", customPVCostStr)
			}

			cost = customPVCost
		} else {
			cost = result.Values[0].Value
		}

		key := DiskIdentifier{cluster, name}
		if _, ok := diskMap[key]; !ok {
			diskMap[key] = &Disk{
				Cluster:   cluster,
				Name:      name,
				Breakdown: &ClusterCostsBreakdown{},
			}
		}

		diskMap[key].Cost = cost * (diskMap[key].Bytes / 1024 / 1024 / 1024) * (diskMap[key].Minutes / 60)
		providerID, _ := result.GetString("provider_id") // just put the providerID set up here, it's the simplest query.
		if providerID != "" {
			diskMap[key].ProviderID = provider.ParsePVID(providerID)
		}
	}

	for _, result := range resPVUsedAvg {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		claimName, err := result.GetString("persistentvolumeclaim")
		if err != nil {
			log.Debugf("ClusterDisks: pv usage data missing persistentvolumeclaim")
			continue
		}
		claimNamespace, err := result.GetString("namespace")
		if err != nil {
			log.Debugf("ClusterDisks: pv usage data missing namespace")
			continue
		}

		var volumeName string

		for _, thatRes := range resPVCInfo {

			thatCluster, err := thatRes.GetString(env.GetPromClusterLabel())
			if err != nil {
				thatCluster = env.GetClusterID()
			}

			thatVolumeName, err := thatRes.GetString("volumename")
			if err != nil {
				log.Debugf("ClusterDisks: pv claim data missing volumename")
				continue
			}

			thatClaimName, err := thatRes.GetString("persistentvolumeclaim")
			if err != nil {
				log.Debugf("ClusterDisks: pv claim data missing persistentvolumeclaim")
				continue
			}

			thatClaimNamespace, err := thatRes.GetString("namespace")
			if err != nil {
				log.Debugf("ClusterDisks: pv claim data missing namespace")
				continue
			}

			if cluster == thatCluster && claimName == thatClaimName && claimNamespace == thatClaimNamespace {
				volumeName = thatVolumeName
			}
		}

		usage := result.Values[0].Value

		key := DiskIdentifier{cluster, volumeName}

		if _, ok := diskMap[key]; !ok {
			diskMap[key] = &Disk{
				Cluster:   cluster,
				Name:      volumeName,
				Breakdown: &ClusterCostsBreakdown{},
			}
		}
		diskMap[key].BytesUsedAvgPtr = &usage
	}

	for _, result := range resPVUsedMax {
		cluster, err := result.GetString(env.GetPromClusterLabel())
		if err != nil {
			cluster = env.GetClusterID()
		}

		claimName, err := result.GetString("persistentvolumeclaim")
		if err != nil {
			log.Debugf("ClusterDisks: pv usage data missing persistentvolumeclaim")
			continue
		}

		claimNamespace, err := result.GetString("namespace")
		if err != nil {
			log.Debugf("ClusterDisks: pv usage data missing namespace")
			continue
		}

		var volumeName string

		for _, thatRes := range resPVCInfo {

			thatCluster, err := thatRes.GetString(env.GetPromClusterLabel())
			if err != nil {
				thatCluster = env.GetClusterID()
			}

			thatVolumeName, err := thatRes.GetString("volumename")
			if err != nil {
				log.Debugf("ClusterDisks: pv claim data missing volumename")
				continue
			}

			thatClaimName, err := thatRes.GetString("persistentvolumeclaim")
			if err != nil {
				log.Debugf("ClusterDisks: pv claim data missing persistentvolumeclaim")
				continue
			}

			thatClaimNamespace, err := thatRes.GetString("namespace")
			if err != nil {
				log.Debugf("ClusterDisks: pv claim data missing namespace")
				continue
			}

			if cluster == thatCluster && claimName == thatClaimName && claimNamespace == thatClaimNamespace {
				volumeName = thatVolumeName
			}
		}

		usage := result.Values[0].Value

		key := DiskIdentifier{cluster, volumeName}

		if _, ok := diskMap[key]; !ok {
			diskMap[key] = &Disk{
				Cluster:   cluster,
				Name:      volumeName,
				Breakdown: &ClusterCostsBreakdown{},
			}
		}
		diskMap[key].BytesUsedMaxPtr = &usage
	}
}

// filterOutLocalPVs removes local Persistent Volumes (PVs) from the given disk map.
// Local PVs are identified by the prefix "local-pv-" in their names, which is the
// convention used by sig-storage-local-static-provisioner.
//
// Parameters:
//   - diskMap: A map of DiskIdentifier to Disk pointers, representing all PVs.
//
// Returns:
//   - A new map of DiskIdentifier to Disk pointers, containing only non-local PVs.
func filterOutLocalPVs(diskMap map[DiskIdentifier]*Disk) map[DiskIdentifier]*Disk {
	nonLocalPVDiskMap := map[DiskIdentifier]*Disk{}
	for key, val := range diskMap {
		if !strings.HasPrefix(key.Name, SIG_STORAGE_LOCAL_PROVISIONER_PREFIX) {
			nonLocalPVDiskMap[key] = val
		}
	}
	return nonLocalPVDiskMap
}