service/ccm,ocm: Fixes and improvements

This commit is contained in:
世界
2026-02-24 19:16:40 +08:00
parent d48236da94
commit 0bc66e5a56
6 changed files with 1201 additions and 287 deletions

View File

@@ -10,6 +10,7 @@ import (
"mime"
"net"
"net/http"
"strconv"
"strings"
"sync"
"time"
@@ -79,6 +80,35 @@ func isHopByHopHeader(header string) bool {
}
}
const (
weeklyWindowSeconds = 604800
weeklyWindowMinutes = weeklyWindowSeconds / 60
)
func parseInt64Header(headers http.Header, headerName string) (int64, bool) {
headerValue := strings.TrimSpace(headers.Get(headerName))
if headerValue == "" {
return 0, false
}
parsedValue, parseError := strconv.ParseInt(headerValue, 10, 64)
if parseError != nil {
return 0, false
}
return parsedValue, true
}
func extractWeeklyCycleHint(headers http.Header) *WeeklyCycleHint {
resetAtUnix, hasResetAt := parseInt64Header(headers, "anthropic-ratelimit-unified-7d-reset")
if !hasResetAt || resetAtUnix <= 0 {
return nil
}
return &WeeklyCycleHint{
WindowMinutes: weeklyWindowMinutes,
ResetAt: time.Unix(resetAtUnix, 0).UTC(),
}
}
type Service struct {
boxService.Adapter
ctx context.Context
@@ -392,6 +422,7 @@ func (s *Service) ServeHTTP(w http.ResponseWriter, r *http.Request) {
}
func (s *Service) handleResponseWithTracking(writer http.ResponseWriter, response *http.Response, requestModel string, anthropicBetaHeader string, messagesCount int, username string) {
weeklyCycleHint := extractWeeklyCycleHint(response.Header)
mediaType, _, err := mime.ParseMediaType(response.Header.Get("Content-Type"))
isStreaming := err == nil && mediaType == "text/event-stream"
@@ -417,7 +448,7 @@ func (s *Service) handleResponseWithTracking(writer http.ResponseWriter, respons
if usage.InputTokens > 0 || usage.OutputTokens > 0 {
if responseModel != "" {
contextWindow := detectContextWindow(anthropicBetaHeader, usage.InputTokens)
s.usageTracker.AddUsage(
s.usageTracker.AddUsageWithCycleHint(
responseModel,
contextWindow,
messagesCount,
@@ -425,7 +456,11 @@ func (s *Service) handleResponseWithTracking(writer http.ResponseWriter, respons
usage.OutputTokens,
usage.CacheReadInputTokens,
usage.CacheCreationInputTokens,
usage.CacheCreation.Ephemeral5mInputTokens,
usage.CacheCreation.Ephemeral1hInputTokens,
username,
time.Now(),
weeklyCycleHint,
)
}
}
@@ -485,6 +520,8 @@ func (s *Service) handleResponseWithTracking(writer http.ResponseWriter, respons
accumulatedUsage.InputTokens = messageStart.Message.Usage.InputTokens
accumulatedUsage.CacheReadInputTokens = messageStart.Message.Usage.CacheReadInputTokens
accumulatedUsage.CacheCreationInputTokens = messageStart.Message.Usage.CacheCreationInputTokens
accumulatedUsage.CacheCreation.Ephemeral5mInputTokens = messageStart.Message.Usage.CacheCreation.Ephemeral5mInputTokens
accumulatedUsage.CacheCreation.Ephemeral1hInputTokens = messageStart.Message.Usage.CacheCreation.Ephemeral1hInputTokens
}
case "message_delta":
messageDelta := event.AsMessageDelta()
@@ -511,7 +548,7 @@ func (s *Service) handleResponseWithTracking(writer http.ResponseWriter, respons
if accumulatedUsage.InputTokens > 0 || accumulatedUsage.OutputTokens > 0 {
if responseModel != "" {
contextWindow := detectContextWindow(anthropicBetaHeader, accumulatedUsage.InputTokens)
s.usageTracker.AddUsage(
s.usageTracker.AddUsageWithCycleHint(
responseModel,
contextWindow,
messagesCount,
@@ -519,7 +556,11 @@ func (s *Service) handleResponseWithTracking(writer http.ResponseWriter, respons
accumulatedUsage.OutputTokens,
accumulatedUsage.CacheReadInputTokens,
accumulatedUsage.CacheCreationInputTokens,
accumulatedUsage.CacheCreation.Ephemeral5mInputTokens,
accumulatedUsage.CacheCreation.Ephemeral1hInputTokens,
username,
time.Now(),
weeklyCycleHint,
)
}
}

View File

@@ -2,6 +2,7 @@ package ccm
import (
"encoding/json"
"fmt"
"math"
"os"
"regexp"
@@ -13,17 +14,20 @@ import (
)
type UsageStats struct {
RequestCount int `json:"request_count"`
MessagesCount int `json:"messages_count"`
InputTokens int64 `json:"input_tokens"`
OutputTokens int64 `json:"output_tokens"`
CacheReadInputTokens int64 `json:"cache_read_input_tokens"`
CacheCreationInputTokens int64 `json:"cache_creation_input_tokens"`
RequestCount int `json:"request_count"`
MessagesCount int `json:"messages_count"`
InputTokens int64 `json:"input_tokens"`
OutputTokens int64 `json:"output_tokens"`
CacheReadInputTokens int64 `json:"cache_read_input_tokens"`
CacheCreationInputTokens int64 `json:"cache_creation_input_tokens"`
CacheCreation5MinuteInputTokens int64 `json:"cache_creation_5m_input_tokens,omitempty"`
CacheCreation1HourInputTokens int64 `json:"cache_creation_1h_input_tokens,omitempty"`
}
type CostCombination struct {
Model string `json:"model"`
ContextWindow int `json:"context_window"`
WeekStartUnix int64 `json:"week_start_unix,omitempty"`
Total UsageStats `json:"total"`
ByUser map[string]UsageStats `json:"by_user"`
}
@@ -41,18 +45,21 @@ type AggregatedUsage struct {
}
type UsageStatsJSON struct {
RequestCount int `json:"request_count"`
MessagesCount int `json:"messages_count"`
InputTokens int64 `json:"input_tokens"`
OutputTokens int64 `json:"output_tokens"`
CacheReadInputTokens int64 `json:"cache_read_input_tokens"`
CacheCreationInputTokens int64 `json:"cache_creation_input_tokens"`
CostUSD float64 `json:"cost_usd"`
RequestCount int `json:"request_count"`
MessagesCount int `json:"messages_count"`
InputTokens int64 `json:"input_tokens"`
OutputTokens int64 `json:"output_tokens"`
CacheReadInputTokens int64 `json:"cache_read_input_tokens"`
CacheCreationInputTokens int64 `json:"cache_creation_input_tokens"`
CacheCreation5MinuteInputTokens int64 `json:"cache_creation_5m_input_tokens,omitempty"`
CacheCreation1HourInputTokens int64 `json:"cache_creation_1h_input_tokens,omitempty"`
CostUSD float64 `json:"cost_usd"`
}
type CostCombinationJSON struct {
Model string `json:"model"`
ContextWindow int `json:"context_window"`
WeekStartUnix int64 `json:"week_start_unix,omitempty"`
Total UsageStatsJSON `json:"total"`
ByUser map[string]UsageStatsJSON `json:"by_user"`
}
@@ -60,6 +67,7 @@ type CostCombinationJSON struct {
type CostsSummaryJSON struct {
TotalUSD float64 `json:"total_usd"`
ByUser map[string]float64 `json:"by_user"`
ByWeek map[string]float64 `json:"by_week,omitempty"`
}
type AggregatedUsageJSON struct {
@@ -68,11 +76,17 @@ type AggregatedUsageJSON struct {
Combinations []CostCombinationJSON `json:"combinations"`
}
type WeeklyCycleHint struct {
WindowMinutes int64
ResetAt time.Time
}
type ModelPricing struct {
InputPrice float64
OutputPrice float64
CacheReadPrice float64
CacheWritePrice float64
InputPrice float64
OutputPrice float64
CacheReadPrice float64
CacheWritePrice5Minute float64
CacheWritePrice1Hour float64
}
type modelFamily struct {
@@ -82,143 +96,205 @@ type modelFamily struct {
}
var (
opus4Pricing = ModelPricing{
InputPrice: 15.0,
OutputPrice: 75.0,
CacheReadPrice: 1.5,
CacheWritePrice: 18.75,
opus46StandardPricing = ModelPricing{
InputPrice: 5.0,
OutputPrice: 25.0,
CacheReadPrice: 0.5,
CacheWritePrice5Minute: 6.25,
CacheWritePrice1Hour: 10.0,
}
sonnet4StandardPricing = ModelPricing{
InputPrice: 3.0,
OutputPrice: 15.0,
CacheReadPrice: 0.3,
CacheWritePrice: 3.75,
}
sonnet4PremiumPricing = ModelPricing{
InputPrice: 6.0,
OutputPrice: 22.5,
CacheReadPrice: 0.6,
CacheWritePrice: 7.5,
}
haiku4Pricing = ModelPricing{
InputPrice: 1.0,
OutputPrice: 5.0,
CacheReadPrice: 0.1,
CacheWritePrice: 1.25,
}
haiku35Pricing = ModelPricing{
InputPrice: 0.8,
OutputPrice: 4.0,
CacheReadPrice: 0.08,
CacheWritePrice: 1.0,
}
sonnet35Pricing = ModelPricing{
InputPrice: 3.0,
OutputPrice: 15.0,
CacheReadPrice: 0.3,
CacheWritePrice: 3.75,
opus46PremiumPricing = ModelPricing{
InputPrice: 10.0,
OutputPrice: 37.5,
CacheReadPrice: 1.0,
CacheWritePrice5Minute: 12.5,
CacheWritePrice1Hour: 20.0,
}
opus45Pricing = ModelPricing{
InputPrice: 5.0,
OutputPrice: 25.0,
CacheReadPrice: 0.5,
CacheWritePrice: 6.25,
InputPrice: 5.0,
OutputPrice: 25.0,
CacheReadPrice: 0.5,
CacheWritePrice5Minute: 6.25,
CacheWritePrice1Hour: 10.0,
}
opus4Pricing = ModelPricing{
InputPrice: 15.0,
OutputPrice: 75.0,
CacheReadPrice: 1.5,
CacheWritePrice5Minute: 18.75,
CacheWritePrice1Hour: 30.0,
}
sonnet46StandardPricing = ModelPricing{
InputPrice: 3.0,
OutputPrice: 15.0,
CacheReadPrice: 0.3,
CacheWritePrice5Minute: 3.75,
CacheWritePrice1Hour: 6.0,
}
sonnet46PremiumPricing = ModelPricing{
InputPrice: 6.0,
OutputPrice: 22.5,
CacheReadPrice: 0.6,
CacheWritePrice5Minute: 7.5,
CacheWritePrice1Hour: 12.0,
}
sonnet45StandardPricing = ModelPricing{
InputPrice: 3.0,
OutputPrice: 15.0,
CacheReadPrice: 0.3,
CacheWritePrice: 3.75,
InputPrice: 3.0,
OutputPrice: 15.0,
CacheReadPrice: 0.3,
CacheWritePrice5Minute: 3.75,
CacheWritePrice1Hour: 6.0,
}
sonnet45PremiumPricing = ModelPricing{
InputPrice: 6.0,
OutputPrice: 22.5,
CacheReadPrice: 0.6,
CacheWritePrice: 7.5,
InputPrice: 6.0,
OutputPrice: 22.5,
CacheReadPrice: 0.6,
CacheWritePrice5Minute: 7.5,
CacheWritePrice1Hour: 12.0,
}
sonnet4StandardPricing = ModelPricing{
InputPrice: 3.0,
OutputPrice: 15.0,
CacheReadPrice: 0.3,
CacheWritePrice5Minute: 3.75,
CacheWritePrice1Hour: 6.0,
}
sonnet4PremiumPricing = ModelPricing{
InputPrice: 6.0,
OutputPrice: 22.5,
CacheReadPrice: 0.6,
CacheWritePrice5Minute: 7.5,
CacheWritePrice1Hour: 12.0,
}
sonnet37Pricing = ModelPricing{
InputPrice: 3.0,
OutputPrice: 15.0,
CacheReadPrice: 0.3,
CacheWritePrice5Minute: 3.75,
CacheWritePrice1Hour: 6.0,
}
sonnet35Pricing = ModelPricing{
InputPrice: 3.0,
OutputPrice: 15.0,
CacheReadPrice: 0.3,
CacheWritePrice5Minute: 3.75,
CacheWritePrice1Hour: 6.0,
}
haiku45Pricing = ModelPricing{
InputPrice: 1.0,
OutputPrice: 5.0,
CacheReadPrice: 0.1,
CacheWritePrice: 1.25,
InputPrice: 1.0,
OutputPrice: 5.0,
CacheReadPrice: 0.1,
CacheWritePrice5Minute: 1.25,
CacheWritePrice1Hour: 2.0,
}
haiku4Pricing = ModelPricing{
InputPrice: 1.0,
OutputPrice: 5.0,
CacheReadPrice: 0.1,
CacheWritePrice5Minute: 1.25,
CacheWritePrice1Hour: 2.0,
}
haiku35Pricing = ModelPricing{
InputPrice: 0.8,
OutputPrice: 4.0,
CacheReadPrice: 0.08,
CacheWritePrice5Minute: 1.0,
CacheWritePrice1Hour: 1.6,
}
haiku3Pricing = ModelPricing{
InputPrice: 0.25,
OutputPrice: 1.25,
CacheReadPrice: 0.03,
CacheWritePrice: 0.3,
InputPrice: 0.25,
OutputPrice: 1.25,
CacheReadPrice: 0.03,
CacheWritePrice5Minute: 0.3,
CacheWritePrice1Hour: 0.5,
}
opus3Pricing = ModelPricing{
InputPrice: 15.0,
OutputPrice: 75.0,
CacheReadPrice: 1.5,
CacheWritePrice: 18.75,
InputPrice: 15.0,
OutputPrice: 75.0,
CacheReadPrice: 1.5,
CacheWritePrice5Minute: 18.75,
CacheWritePrice1Hour: 30.0,
}
modelFamilies = []modelFamily{
{
pattern: regexp.MustCompile(`^claude-opus-4-5-`),
pattern: regexp.MustCompile(`^claude-opus-4-6(?:-|$)`),
standardPricing: opus46StandardPricing,
premiumPricing: &opus46PremiumPricing,
},
{
pattern: regexp.MustCompile(`^claude-opus-4-5(?:-|$)`),
standardPricing: opus45Pricing,
premiumPricing: nil,
},
{
pattern: regexp.MustCompile(`^claude-(?:opus-4-|4-opus-|opus-4-1-)`),
pattern: regexp.MustCompile(`^claude-(?:opus-4(?:-|$)|4-opus-)`),
standardPricing: opus4Pricing,
premiumPricing: nil,
},
{
pattern: regexp.MustCompile(`^claude-(?:opus-3-|3-opus-)`),
pattern: regexp.MustCompile(`^claude-(?:opus-3(?:-|$)|3-opus-)`),
standardPricing: opus3Pricing,
premiumPricing: nil,
},
{
pattern: regexp.MustCompile(`^claude-(?:sonnet-4-5-|4-5-sonnet-)`),
pattern: regexp.MustCompile(`^claude-(?:sonnet-4-6(?:-|$)|4-6-sonnet-)`),
standardPricing: sonnet46StandardPricing,
premiumPricing: &sonnet46PremiumPricing,
},
{
pattern: regexp.MustCompile(`^claude-(?:sonnet-4-5(?:-|$)|4-5-sonnet-)`),
standardPricing: sonnet45StandardPricing,
premiumPricing: &sonnet45PremiumPricing,
},
{
pattern: regexp.MustCompile(`^claude-3-7-sonnet-`),
pattern: regexp.MustCompile(`^claude-(?:sonnet-4(?:-|$)|4-sonnet-)`),
standardPricing: sonnet4StandardPricing,
premiumPricing: &sonnet4PremiumPricing,
},
{
pattern: regexp.MustCompile(`^claude-(?:sonnet-4-|4-sonnet-)`),
standardPricing: sonnet4StandardPricing,
premiumPricing: &sonnet4PremiumPricing,
pattern: regexp.MustCompile(`^claude-3-7-sonnet(?:-|$)`),
standardPricing: sonnet37Pricing,
premiumPricing: nil,
},
{
pattern: regexp.MustCompile(`^claude-3-5-sonnet-`),
pattern: regexp.MustCompile(`^claude-3-5-sonnet(?:-|$)`),
standardPricing: sonnet35Pricing,
premiumPricing: nil,
},
{
pattern: regexp.MustCompile(`^claude-(?:haiku-4-5-|4-5-haiku-)`),
pattern: regexp.MustCompile(`^claude-(?:haiku-4-5(?:-|$)|4-5-haiku-)`),
standardPricing: haiku45Pricing,
premiumPricing: nil,
},
{
pattern: regexp.MustCompile(`^claude-haiku-4-`),
pattern: regexp.MustCompile(`^claude-haiku-4(?:-|$)`),
standardPricing: haiku4Pricing,
premiumPricing: nil,
},
{
pattern: regexp.MustCompile(`^claude-3-5-haiku-`),
pattern: regexp.MustCompile(`^claude-3-5-haiku(?:-|$)`),
standardPricing: haiku35Pricing,
premiumPricing: nil,
},
{
pattern: regexp.MustCompile(`^claude-3-haiku-`),
pattern: regexp.MustCompile(`^claude-3-haiku(?:-|$)`),
standardPricing: haiku3Pricing,
premiumPricing: nil,
},
@@ -243,68 +319,211 @@ func getPricing(model string, contextWindow int) ModelPricing {
func calculateCost(stats UsageStats, model string, contextWindow int) float64 {
pricing := getPricing(model, contextWindow)
cacheCreationCost := 0.0
if stats.CacheCreation5MinuteInputTokens > 0 || stats.CacheCreation1HourInputTokens > 0 {
cacheCreationCost = float64(stats.CacheCreation5MinuteInputTokens)*pricing.CacheWritePrice5Minute +
float64(stats.CacheCreation1HourInputTokens)*pricing.CacheWritePrice1Hour
} else {
// Backward compatibility for usage files generated before TTL split tracking.
cacheCreationCost = float64(stats.CacheCreationInputTokens) * pricing.CacheWritePrice5Minute
}
cost := (float64(stats.InputTokens)*pricing.InputPrice +
float64(stats.OutputTokens)*pricing.OutputPrice +
float64(stats.CacheReadInputTokens)*pricing.CacheReadPrice +
float64(stats.CacheCreationInputTokens)*pricing.CacheWritePrice) / 1_000_000
cacheCreationCost) / 1_000_000
return math.Round(cost*100) / 100
}
func roundCost(cost float64) float64 {
return math.Round(cost*100) / 100
}
func normalizeCombinations(combinations []CostCombination) {
for index := range combinations {
if combinations[index].ByUser == nil {
combinations[index].ByUser = make(map[string]UsageStats)
}
}
}
func addUsageToCombinations(
combinations *[]CostCombination,
model string,
contextWindow int,
weekStartUnix int64,
messagesCount int,
inputTokens, outputTokens, cacheReadTokens, cacheCreationTokens, cacheCreation5MinuteTokens, cacheCreation1HourTokens int64,
user string,
) {
var matchedCombination *CostCombination
for index := range *combinations {
combination := &(*combinations)[index]
if combination.Model == model && combination.ContextWindow == contextWindow && combination.WeekStartUnix == weekStartUnix {
matchedCombination = combination
break
}
}
if matchedCombination == nil {
newCombination := CostCombination{
Model: model,
ContextWindow: contextWindow,
WeekStartUnix: weekStartUnix,
Total: UsageStats{},
ByUser: make(map[string]UsageStats),
}
*combinations = append(*combinations, newCombination)
matchedCombination = &(*combinations)[len(*combinations)-1]
}
if cacheCreationTokens == 0 {
cacheCreationTokens = cacheCreation5MinuteTokens + cacheCreation1HourTokens
}
matchedCombination.Total.RequestCount++
matchedCombination.Total.MessagesCount += messagesCount
matchedCombination.Total.InputTokens += inputTokens
matchedCombination.Total.OutputTokens += outputTokens
matchedCombination.Total.CacheReadInputTokens += cacheReadTokens
matchedCombination.Total.CacheCreationInputTokens += cacheCreationTokens
matchedCombination.Total.CacheCreation5MinuteInputTokens += cacheCreation5MinuteTokens
matchedCombination.Total.CacheCreation1HourInputTokens += cacheCreation1HourTokens
if user != "" {
userStats := matchedCombination.ByUser[user]
userStats.RequestCount++
userStats.MessagesCount += messagesCount
userStats.InputTokens += inputTokens
userStats.OutputTokens += outputTokens
userStats.CacheReadInputTokens += cacheReadTokens
userStats.CacheCreationInputTokens += cacheCreationTokens
userStats.CacheCreation5MinuteInputTokens += cacheCreation5MinuteTokens
userStats.CacheCreation1HourInputTokens += cacheCreation1HourTokens
matchedCombination.ByUser[user] = userStats
}
}
func buildCombinationJSON(combinations []CostCombination, aggregateUserCosts map[string]float64) ([]CostCombinationJSON, float64) {
result := make([]CostCombinationJSON, len(combinations))
var totalCost float64
for index, combination := range combinations {
combinationTotalCost := calculateCost(combination.Total, combination.Model, combination.ContextWindow)
totalCost += combinationTotalCost
combinationJSON := CostCombinationJSON{
Model: combination.Model,
ContextWindow: combination.ContextWindow,
WeekStartUnix: combination.WeekStartUnix,
Total: UsageStatsJSON{
RequestCount: combination.Total.RequestCount,
MessagesCount: combination.Total.MessagesCount,
InputTokens: combination.Total.InputTokens,
OutputTokens: combination.Total.OutputTokens,
CacheReadInputTokens: combination.Total.CacheReadInputTokens,
CacheCreationInputTokens: combination.Total.CacheCreationInputTokens,
CacheCreation5MinuteInputTokens: combination.Total.CacheCreation5MinuteInputTokens,
CacheCreation1HourInputTokens: combination.Total.CacheCreation1HourInputTokens,
CostUSD: combinationTotalCost,
},
ByUser: make(map[string]UsageStatsJSON),
}
for user, userStats := range combination.ByUser {
userCost := calculateCost(userStats, combination.Model, combination.ContextWindow)
if aggregateUserCosts != nil {
aggregateUserCosts[user] += userCost
}
combinationJSON.ByUser[user] = UsageStatsJSON{
RequestCount: userStats.RequestCount,
MessagesCount: userStats.MessagesCount,
InputTokens: userStats.InputTokens,
OutputTokens: userStats.OutputTokens,
CacheReadInputTokens: userStats.CacheReadInputTokens,
CacheCreationInputTokens: userStats.CacheCreationInputTokens,
CacheCreation5MinuteInputTokens: userStats.CacheCreation5MinuteInputTokens,
CacheCreation1HourInputTokens: userStats.CacheCreation1HourInputTokens,
CostUSD: userCost,
}
}
result[index] = combinationJSON
}
return result, roundCost(totalCost)
}
func formatUTCOffsetLabel(timestamp time.Time) string {
_, offsetSeconds := timestamp.Zone()
sign := "+"
if offsetSeconds < 0 {
sign = "-"
offsetSeconds = -offsetSeconds
}
offsetHours := offsetSeconds / 3600
offsetMinutes := (offsetSeconds % 3600) / 60
if offsetMinutes == 0 {
return fmt.Sprintf("UTC%s%d", sign, offsetHours)
}
return fmt.Sprintf("UTC%s%d:%02d", sign, offsetHours, offsetMinutes)
}
func formatWeekStartKey(cycleStartAt time.Time) string {
localCycleStart := cycleStartAt.In(time.Local)
return fmt.Sprintf("%s %s", localCycleStart.Format("2006-01-02 15:04:05"), formatUTCOffsetLabel(localCycleStart))
}
func buildByWeekCost(combinations []CostCombination) map[string]float64 {
byWeek := make(map[string]float64)
for _, combination := range combinations {
if combination.WeekStartUnix <= 0 {
continue
}
weekStartAt := time.Unix(combination.WeekStartUnix, 0).UTC()
weekKey := formatWeekStartKey(weekStartAt)
byWeek[weekKey] += calculateCost(combination.Total, combination.Model, combination.ContextWindow)
}
for weekKey, weekCost := range byWeek {
byWeek[weekKey] = roundCost(weekCost)
}
return byWeek
}
func deriveWeekStartUnix(cycleHint *WeeklyCycleHint) int64 {
if cycleHint == nil || cycleHint.WindowMinutes <= 0 || cycleHint.ResetAt.IsZero() {
return 0
}
windowDuration := time.Duration(cycleHint.WindowMinutes) * time.Minute
return cycleHint.ResetAt.UTC().Add(-windowDuration).Unix()
}
func (u *AggregatedUsage) ToJSON() *AggregatedUsageJSON {
u.mutex.Lock()
defer u.mutex.Unlock()
result := &AggregatedUsageJSON{
LastUpdated: u.LastUpdated,
Combinations: make([]CostCombinationJSON, len(u.Combinations)),
LastUpdated: u.LastUpdated,
Costs: CostsSummaryJSON{
TotalUSD: 0,
ByUser: make(map[string]float64),
ByWeek: make(map[string]float64),
},
}
for i, combo := range u.Combinations {
totalCost := calculateCost(combo.Total, combo.Model, combo.ContextWindow)
globalCombinationsJSON, totalCost := buildCombinationJSON(u.Combinations, result.Costs.ByUser)
result.Combinations = globalCombinationsJSON
result.Costs.TotalUSD = totalCost
result.Costs.ByWeek = buildByWeekCost(u.Combinations)
result.Costs.TotalUSD += totalCost
comboJSON := CostCombinationJSON{
Model: combo.Model,
ContextWindow: combo.ContextWindow,
Total: UsageStatsJSON{
RequestCount: combo.Total.RequestCount,
MessagesCount: combo.Total.MessagesCount,
InputTokens: combo.Total.InputTokens,
OutputTokens: combo.Total.OutputTokens,
CacheReadInputTokens: combo.Total.CacheReadInputTokens,
CacheCreationInputTokens: combo.Total.CacheCreationInputTokens,
CostUSD: totalCost,
},
ByUser: make(map[string]UsageStatsJSON),
}
for user, userStats := range combo.ByUser {
userCost := calculateCost(userStats, combo.Model, combo.ContextWindow)
result.Costs.ByUser[user] += userCost
comboJSON.ByUser[user] = UsageStatsJSON{
RequestCount: userStats.RequestCount,
MessagesCount: userStats.MessagesCount,
InputTokens: userStats.InputTokens,
OutputTokens: userStats.OutputTokens,
CacheReadInputTokens: userStats.CacheReadInputTokens,
CacheCreationInputTokens: userStats.CacheCreationInputTokens,
CostUSD: userCost,
}
}
result.Combinations[i] = comboJSON
if len(result.Costs.ByWeek) == 0 {
result.Costs.ByWeek = nil
}
result.Costs.TotalUSD = math.Round(result.Costs.TotalUSD*100) / 100
for user, cost := range result.Costs.ByUser {
result.Costs.ByUser[user] = math.Round(cost*100) / 100
result.Costs.ByUser[user] = roundCost(cost)
}
return result
@@ -314,6 +533,9 @@ func (u *AggregatedUsage) Load() error {
u.mutex.Lock()
defer u.mutex.Unlock()
u.LastUpdated = time.Time{}
u.Combinations = nil
data, err := os.ReadFile(u.filePath)
if err != nil {
if os.IsNotExist(err) {
@@ -334,12 +556,7 @@ func (u *AggregatedUsage) Load() error {
u.LastUpdated = temp.LastUpdated
u.Combinations = temp.Combinations
for i := range u.Combinations {
if u.Combinations[i].ByUser == nil {
u.Combinations[i].ByUser = make(map[string]UsageStats)
}
}
normalizeCombinations(u.Combinations)
return nil
}
@@ -367,58 +584,42 @@ func (u *AggregatedUsage) Save() error {
return err
}
func (u *AggregatedUsage) AddUsage(model string, contextWindow int, messagesCount int, inputTokens, outputTokens, cacheReadTokens, cacheCreationTokens int64, user string) error {
func (u *AggregatedUsage) AddUsage(
model string,
contextWindow int,
messagesCount int,
inputTokens, outputTokens, cacheReadTokens, cacheCreationTokens, cacheCreation5MinuteTokens, cacheCreation1HourTokens int64,
user string,
) error {
return u.AddUsageWithCycleHint(model, contextWindow, messagesCount, inputTokens, outputTokens, cacheReadTokens, cacheCreationTokens, cacheCreation5MinuteTokens, cacheCreation1HourTokens, user, time.Now(), nil)
}
func (u *AggregatedUsage) AddUsageWithCycleHint(
model string,
contextWindow int,
messagesCount int,
inputTokens, outputTokens, cacheReadTokens, cacheCreationTokens, cacheCreation5MinuteTokens, cacheCreation1HourTokens int64,
user string,
observedAt time.Time,
cycleHint *WeeklyCycleHint,
) error {
if model == "" {
return E.New("model cannot be empty")
}
if contextWindow <= 0 {
return E.New("contextWindow must be positive")
}
if observedAt.IsZero() {
observedAt = time.Now()
}
u.mutex.Lock()
defer u.mutex.Unlock()
u.LastUpdated = time.Now()
u.LastUpdated = observedAt
weekStartUnix := deriveWeekStartUnix(cycleHint)
// Find or create combination
var combo *CostCombination
for i := range u.Combinations {
if u.Combinations[i].Model == model && u.Combinations[i].ContextWindow == contextWindow {
combo = &u.Combinations[i]
break
}
}
if combo == nil {
newCombo := CostCombination{
Model: model,
ContextWindow: contextWindow,
Total: UsageStats{},
ByUser: make(map[string]UsageStats),
}
u.Combinations = append(u.Combinations, newCombo)
combo = &u.Combinations[len(u.Combinations)-1]
}
// Update total stats
combo.Total.RequestCount++
combo.Total.MessagesCount += messagesCount
combo.Total.InputTokens += inputTokens
combo.Total.OutputTokens += outputTokens
combo.Total.CacheReadInputTokens += cacheReadTokens
combo.Total.CacheCreationInputTokens += cacheCreationTokens
// Update per-user stats if user is specified
if user != "" {
userStats := combo.ByUser[user]
userStats.RequestCount++
userStats.MessagesCount += messagesCount
userStats.InputTokens += inputTokens
userStats.OutputTokens += outputTokens
userStats.CacheReadInputTokens += cacheReadTokens
userStats.CacheCreationInputTokens += cacheCreationTokens
combo.ByUser[user] = userStats
}
addUsageToCombinations(&u.Combinations, model, contextWindow, weekStartUnix, messagesCount, inputTokens, outputTokens, cacheReadTokens, cacheCreationTokens, cacheCreation5MinuteTokens, cacheCreation1HourTokens, user)
go u.scheduleSave()