wanwu/sub2api
1
0
Fork 0
mirror of https://gitee.com/wanwujie/sub2api synced 2026-04-19 06:14:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor: isolate claude max response usage simulation by group toggle

Browse Source
This commit is contained in:
erio
2026-02-27 16:14:07 +08:00
parent e71be7e0f1
commit 61ef73cb12
5 changed files with 356 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
backend/internal/service/gateway_service.go
View File

@@ -3709,6 +3709,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
}
// 处理正常响应
ctx = withClaudeMaxResponseRewriteContext(ctx, c, parsed)
var usage *ClaudeUsage
var firstTokenMs *int
var clientDisconnect bool
@@ -5105,6 +5106,7 @@ func (s *GatewayService) handleStreamingResponse(ctx context.Context, resp *http
needModelReplace := originalModel != mappedModel
clientDisconnected := false // 客户端断开标志，断开后继续读取上游以获取完整usage
skipAccountTTLOverride := false
pendingEventLines := make([]string, 0, 4)
@@ -5164,17 +5166,25 @@ func (s *GatewayService) handleStreamingResponse(ctx context.Context, resp *http
if msg, ok := event["message"].(map[string]any); ok {
if u, ok := msg["usage"].(map[string]any); ok {
reconcileCachedTokens(u)
claudeMaxOutcome := applyClaudeMaxSimulationToUsageJSONMap(ctx, u, originalModel, account.ID)
if claudeMaxOutcome.Simulated {
skipAccountTTLOverride = true
}
}
}
}
if eventType == "message_delta" {
if u, ok := event["usage"].(map[string]any); ok {
reconcileCachedTokens(u)
claudeMaxOutcome := applyClaudeMaxSimulationToUsageJSONMap(ctx, u, originalModel, account.ID)
if claudeMaxOutcome.Simulated {
skipAccountTTLOverride = true
}
}
}
// Cache TTL Override: 重写 SSE 事件中的 cache_creation 分类
if account.IsCacheTTLOverrideEnabled() {
if account.IsCacheTTLOverrideEnabled() && !skipAccountTTLOverride {
overrideTarget := account.GetCacheTTLOverrideTarget()
if eventType == "message_start" {
if msg, ok := event["message"].(map[string]any); ok {
@@ -5465,8 +5475,13 @@ func (s *GatewayService) handleNonStreamingResponse(ctx context.Context, resp *h
}
}
claudeMaxOutcome := applyClaudeMaxSimulationToUsage(ctx, &response.Usage, originalModel, account.ID)
if claudeMaxOutcome.Simulated {
body = rewriteClaudeUsageJSONBytes(body, response.Usage)
}
// Cache TTL Override: 重写 non-streaming 响应中的 cache_creation 分类
if account.IsCacheTTLOverrideEnabled() {
if account.IsCacheTTLOverrideEnabled() && !claudeMaxOutcome.Simulated && !claudeMaxOutcome.ForcedCache1H {
overrideTarget := account.GetCacheTTLOverrideTarget()
if applyCacheTTLOverride(&response.Usage, overrideTarget) {
// 同步更新 body JSON 中的嵌套 cache_creation 对象
@@ -5608,9 +5623,12 @@ func (s *GatewayService) RecordUsage(ctx context.Context, input *RecordUsageInpu
result.Usage.InputTokens = 0
}
// Claude Max cache billing policy (group-level): force existing cache creation to 1h,
// otherwise simulate projection only when request carries cache signals.
claudeMaxOutcome := applyClaudeMaxCacheBillingPolicy(input)
// Claude Max cache billing policy (group-level): RecordUsage only checks outcome.
var apiKeyGroup *Group
if apiKey != nil {
apiKeyGroup = apiKey.Group
}
claudeMaxOutcome := detectClaudeMaxCacheBillingOutcomeForUsage(result.Usage, input.ParsedRequest, apiKeyGroup, result.Model)
simulatedClaudeMax := claudeMaxOutcome.Simulated
forcedClaudeMax1H := claudeMaxOutcome.ForcedCache1H