Files
sub2api/backend/internal/handler/openai_gateway_count_tokens.go
T
shaw 0fd2e9216d fix(scheduler): 修复 OpenAI 高级调度器审计发现的正确性与性能问题
针对 #3692 合并后审计发现的问题集中修复:

- previous_response_id 剥离条件改为按 call_id 全覆盖校验,
  部分可重建的工具续链不再被误剥离(不受开关门控的行为回归)
- 粘性加权回退路径补分组归属校验并清理失效绑定,杜绝跨分组账号泄漏
- 账号列表页:无 OpenAI 账号时跳过分数计算、过滤池限定 openai 平台、
  负载批查合并为账号并集一次查询,消除全表扫描与 Redis N+1
- 订阅优先模式下常规池不可用时回退订阅池等待计划,
  busy-but-waitable 的订阅账号不再导致请求硬失败
- TopK/权重 DB 覆盖显式受总开关门控,与兄弟子开关语义一致
- 前端未分组 OpenAI 账号回退展示基础分,不再显示 "-"
- ListAllWithFilters 等能力正式进入 AccountRepository/AdminService 接口,
  移除匿名接口断言与静默降级;负载批查失败补 warn 日志
- SelectAccountWithSchedulerForCapability 增加显式 previousResponseCanMove
  参数,移除 "previous_response_can_move" 魔法字符串哨兵
- 设置写入路径补"基础权重不得全为零"聚合校验;
  运行时设置批量读取失败的降级路径覆盖全部键并留痕
2026-07-06 11:43:16 +08:00

147 lines
5.3 KiB
Go

package handler
import (
"net/http"
"strconv"
"time"
"github.com/Wei-Shaw/sub2api/internal/domain"
pkghttputil "github.com/Wei-Shaw/sub2api/internal/pkg/httputil"
middleware2 "github.com/Wei-Shaw/sub2api/internal/server/middleware"
"github.com/Wei-Shaw/sub2api/internal/service"
"github.com/gin-gonic/gin"
"go.uber.org/zap"
)
// CountTokens handles Anthropic-compatible POST /v1/messages/count_tokens for OpenAI groups.
// It validates billing and routes to an OpenAI token-count bridge without taking concurrency slots
// or recording usage.
func (h *OpenAIGatewayHandler) CountTokens(c *gin.Context) {
apiKey, ok := middleware2.GetAPIKeyFromContext(c)
if !ok {
h.anthropicErrorResponse(c, http.StatusUnauthorized, "authentication_error", "Invalid API key")
return
}
subject, ok := middleware2.GetAuthSubjectFromContext(c)
if !ok {
h.anthropicErrorResponse(c, http.StatusInternalServerError, "api_error", "User context not found")
return
}
reqLog := requestLogger(
c,
"handler.openai_gateway.count_tokens",
zap.Int64("user_id", subject.UserID),
zap.Int64("api_key_id", apiKey.ID),
zap.Any("group_id", apiKey.GroupID),
)
if apiKey.Group != nil && !apiKey.Group.AllowMessagesDispatch {
h.anthropicErrorResponse(c, http.StatusForbidden, "permission_error",
"This group does not allow /v1/messages dispatch")
return
}
if !h.ensureResponsesDependencies(c, reqLog) {
return
}
body, err := pkghttputil.ReadRequestBodyWithPrealloc(c.Request)
if err != nil {
if maxErr, ok := extractMaxBytesError(err); ok {
h.anthropicErrorResponse(c, http.StatusRequestEntityTooLarge, "invalid_request_error", buildBodyTooLargeMessage(maxErr.Limit))
return
}
h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to read request body")
return
}
if len(body) == 0 {
h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Request body is empty")
return
}
bodyRef := service.NewRequestBodyRef(body)
parsedReq, err := service.ParseGatewayRequest(bodyRef, domain.PlatformAnthropic)
if err != nil {
h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to parse request body")
return
}
if parsedReq.Model == "" {
h.anthropicErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "model is required")
return
}
reqModel := parsedReq.Model
routingModel := service.NormalizeOpenAICompatRequestedModel(reqModel)
preferredMappedModel := resolveOpenAIMessagesDispatchMappedModel(apiKey, reqModel)
reqLog = reqLog.With(zap.String("model", reqModel), zap.Bool("stream", parsedReq.Stream))
setOpsRequestContext(c, reqModel, false)
setOpsEndpointContext(c, "", int16(service.RequestTypeFromLegacy(false, false)))
channelMapping, _ := h.gatewayService.ResolveChannelMappingAndRestrict(c.Request.Context(), apiKey.GroupID, reqModel)
mappedBodyForMessages := newOpenAIModelMappedBodyCache(body, h.gatewayService.ReplaceModelInBody)
subscription, _ := middleware2.GetSubscriptionFromContext(c)
if err := h.billingCacheService.CheckBillingEligibility(c.Request.Context(), apiKey.User, apiKey, apiKey.Group, subscription, service.QuotaPlatform(c.Request.Context(), apiKey)); err != nil {
reqLog.Info("openai_count_tokens.billing_eligibility_check_failed", zap.Error(err))
status, code, message, retryAfter := billingErrorDetails(err)
if retryAfter > 0 {
c.Header("Retry-After", strconv.Itoa(retryAfter))
}
h.anthropicErrorResponse(c, status, code, message)
return
}
requestStart := time.Now()
sessionHash := h.gatewayService.GenerateSessionHash(c, body)
currentRoutingModel := routingModel
if preferredMappedModel != "" {
currentRoutingModel = preferredMappedModel
}
selection, _, err := h.gatewayService.SelectAccountWithSchedulerForCapability(
c.Request.Context(),
apiKey.GroupID,
"",
sessionHash,
currentRoutingModel,
nil,
service.OpenAIUpstreamTransportAny,
service.OpenAIEndpointCapabilityChatCompletions,
false,
false,
openAICompatibleRequestPlatform(apiKey),
)
service.SetOpsLatencyMs(c, service.OpsAuthLatencyMsKey, time.Since(requestStart).Milliseconds())
if err != nil {
reqLog.Warn("openai_count_tokens.account_select_failed", zap.Error(err))
cls := classifyNoAccountErrorFromGin(c, h.gatewayService, apiKey, currentRoutingModel, reqModel, service.PlatformOpenAI)
if !cls.ModelNotFound {
markOpsRoutingCapacityLimitedIfNoAvailable(c, err)
}
h.anthropicErrorResponse(c, cls.Status, cls.ErrType, cls.Message)
return
}
if selection == nil || selection.Account == nil {
cls := classifyNoAccountErrorFromGin(c, h.gatewayService, apiKey, currentRoutingModel, reqModel, service.PlatformOpenAI)
if !cls.ModelNotFound {
markOpsRoutingCapacityLimited(c)
}
h.anthropicErrorResponse(c, cls.Status, cls.ErrType, cls.Message)
return
}
account := selection.Account
setOpsSelectedAccount(c, account.ID, account.Platform)
if selection.Acquired && selection.ReleaseFunc != nil {
defer selection.ReleaseFunc()
}
forwardBody := mappedBodyForMessages(channelMapping.Mapped, channelMapping.MappedModel)
defaultMappedModel := preferredMappedModel
if err := h.gatewayService.ForwardCountTokensAsAnthropic(c.Request.Context(), c, account, forwardBody, defaultMappedModel); err != nil {
reqLog.Error("openai_count_tokens.forward_failed", zap.Int64("account_id", account.ID), zap.Error(err))
}
}