feat: 添加5h窗口费用控制和会话数量限制

- 支持Anthropic OAuth/SetupToken账号的5h窗口费用阈值控制
- 支持账号级别的并发会话数量限制
- 使用Redis缓存窗口费用(30秒TTL)减少数据库压力
- 费用计算基于标准费用(不含账号倍率)
This commit is contained in:
shaw
2026-01-16 23:36:52 +08:00
parent 04811c00cb
commit 7379423325
21 changed files with 1329 additions and 53 deletions

View File

@@ -185,7 +185,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
lastFailoverStatus := 0
for {
selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionKey, reqModel, failedAccountIDs)
selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionKey, reqModel, failedAccountIDs, "") // Gemini 不使用会话限制
if err != nil {
if len(failedAccountIDs) == 0 {
h.handleStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "No available accounts: "+err.Error(), streamStarted)
@@ -320,7 +320,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
for {
// 选择支持该模型的账号
selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionKey, reqModel, failedAccountIDs)
selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionKey, reqModel, failedAccountIDs, parsedReq.MetadataUserID)
if err != nil {
if len(failedAccountIDs) == 0 {
h.handleStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "No available accounts: "+err.Error(), streamStarted)