Merge pull request #1242 from Ethan0x0000/feat/anthropic-openai-endpoint-compat

支持 Anthropic Responses / Chat Completions 兼容端点并完善会话一致性与错误可观测性

backend/internal/handler/gateway_handler.go

@@ -178,6 +178,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 	c.Request = c.Request.WithContext(service.WithThinkingEnabled(c.Request.Context(), parsedReq.ThinkingEnabled, h.metadataBridgeEnabled()))
 
 	setOpsRequestContext(c, reqModel, reqStream, body)
+	setOpsEndpointContext(c, "", int16(service.RequestTypeFromLegacy(reqStream, false)))
 
 	// 验证 model 必填
 	if reqModel == "" {
@@ -1396,6 +1397,7 @@ func (h *GatewayHandler) CountTokens(c *gin.Context) {
 	}
 
 	setOpsRequestContext(c, parsedReq.Model, parsedReq.Stream, body)
+	setOpsEndpointContext(c, "", int16(service.RequestTypeFromLegacy(parsedReq.Stream, false)))
 
 	// 获取订阅信息（可能为nil）
 	subscription, _ := middleware2.GetSubscriptionFromContext(c)

backend/internal/handler/gateway_handler_chat_completions.go

@@ -0,0 +1,289 @@
+package handler
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"time"
+
+	pkghttputil "github.com/Wei-Shaw/sub2api/internal/pkg/httputil"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/ip"
+	middleware2 "github.com/Wei-Shaw/sub2api/internal/server/middleware"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
+	"go.uber.org/zap"
+)
+
+// ChatCompletions handles OpenAI Chat Completions API endpoint for Anthropic platform groups.
+// POST /v1/chat/completions
+// This converts Chat Completions requests to Anthropic format (via Responses format chain),
+// forwards to Anthropic upstream, and converts responses back to Chat Completions format.
+func (h *GatewayHandler) ChatCompletions(c *gin.Context) {
+	streamStarted := false
+
+	requestStart := time.Now()
+
+	apiKey, ok := middleware2.GetAPIKeyFromContext(c)
+	if !ok {
+		h.chatCompletionsErrorResponse(c, http.StatusUnauthorized, "authentication_error", "Invalid API key")
+		return
+	}
+
+	subject, ok := middleware2.GetAuthSubjectFromContext(c)
+	if !ok {
+		h.chatCompletionsErrorResponse(c, http.StatusInternalServerError, "api_error", "User context not found")
+		return
+	}
+	reqLog := requestLogger(
+		c,
+		"handler.gateway.chat_completions",
+		zap.Int64("user_id", subject.UserID),
+		zap.Int64("api_key_id", apiKey.ID),
+		zap.Any("group_id", apiKey.GroupID),
+	)
+
+	// Read request body
+	body, err := pkghttputil.ReadRequestBodyWithPrealloc(c.Request)
+	if err != nil {
+		if maxErr, ok := extractMaxBytesError(err); ok {
+			h.chatCompletionsErrorResponse(c, http.StatusRequestEntityTooLarge, "invalid_request_error", buildBodyTooLargeMessage(maxErr.Limit))
+			return
+		}
+		h.chatCompletionsErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to read request body")
+		return
+	}
+
+	if len(body) == 0 {
+		h.chatCompletionsErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Request body is empty")
+		return
+	}
+
+	setOpsRequestContext(c, "", false, body)
+
+	// Validate JSON
+	if !gjson.ValidBytes(body) {
+		h.chatCompletionsErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to parse request body")
+		return
+	}
+
+	// Extract model and stream
+	modelResult := gjson.GetBytes(body, "model")
+	if !modelResult.Exists() || modelResult.Type != gjson.String || modelResult.String() == "" {
+		h.chatCompletionsErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "model is required")
+		return
+	}
+	reqModel := modelResult.String()
+	reqStream := gjson.GetBytes(body, "stream").Bool()
+	reqLog = reqLog.With(zap.String("model", reqModel), zap.Bool("stream", reqStream))
+
+	setOpsRequestContext(c, reqModel, reqStream, body)
+	setOpsEndpointContext(c, "", int16(service.RequestTypeFromLegacy(reqStream, false)))
+
+	// Claude Code only restriction
+	if apiKey.Group != nil && apiKey.Group.ClaudeCodeOnly {
+		h.chatCompletionsErrorResponse(c, http.StatusForbidden, "permission_error",
+			"This group is restricted to Claude Code clients (/v1/messages only)")
+		return
+	}
+
+	// Error passthrough binding
+	if h.errorPassthroughService != nil {
+		service.BindErrorPassthroughService(c, h.errorPassthroughService)
+	}
+
+	subscription, _ := middleware2.GetSubscriptionFromContext(c)
+
+	service.SetOpsLatencyMs(c, service.OpsAuthLatencyMsKey, time.Since(requestStart).Milliseconds())
+
+	// 1. Acquire user concurrency slot
+	maxWait := service.CalculateMaxWait(subject.Concurrency)
+	canWait, err := h.concurrencyHelper.IncrementWaitCount(c.Request.Context(), subject.UserID, maxWait)
+	waitCounted := false
+	if err != nil {
+		reqLog.Warn("gateway.cc.user_wait_counter_increment_failed", zap.Error(err))
+	} else if !canWait {
+		h.chatCompletionsErrorResponse(c, http.StatusTooManyRequests, "rate_limit_error", "Too many pending requests, please retry later")
+		return
+	}
+	if err == nil && canWait {
+		waitCounted = true
+	}
+	defer func() {
+		if waitCounted {
+			h.concurrencyHelper.DecrementWaitCount(c.Request.Context(), subject.UserID)
+		}
+	}()
+
+	userReleaseFunc, err := h.concurrencyHelper.AcquireUserSlotWithWait(c, subject.UserID, subject.Concurrency, reqStream, &streamStarted)
+	if err != nil {
+		reqLog.Warn("gateway.cc.user_slot_acquire_failed", zap.Error(err))
+		h.handleConcurrencyError(c, err, "user", streamStarted)
+		return
+	}
+	if waitCounted {
+		h.concurrencyHelper.DecrementWaitCount(c.Request.Context(), subject.UserID)
+		waitCounted = false
+	}
+	userReleaseFunc = wrapReleaseOnDone(c.Request.Context(), userReleaseFunc)
+	if userReleaseFunc != nil {
+		defer userReleaseFunc()
+	}
+
+	// 2. Re-check billing
+	if err := h.billingCacheService.CheckBillingEligibility(c.Request.Context(), apiKey.User, apiKey, apiKey.Group, subscription); err != nil {
+		reqLog.Info("gateway.cc.billing_check_failed", zap.Error(err))
+		status, code, message := billingErrorDetails(err)
+		h.chatCompletionsErrorResponse(c, status, code, message)
+		return
+	}
+
+	// Parse request for session hash
+	parsedReq, _ := service.ParseGatewayRequest(body, "chat_completions")
+	if parsedReq == nil {
+		parsedReq = &service.ParsedRequest{Model: reqModel, Stream: reqStream, Body: body}
+	}
+	parsedReq.SessionContext = &service.SessionContext{
+		ClientIP:  ip.GetClientIP(c),
+		UserAgent: c.GetHeader("User-Agent"),
+		APIKeyID:  apiKey.ID,
+	}
+	sessionHash := h.gatewayService.GenerateSessionHash(parsedReq)
+
+	// 3. Account selection + failover loop
+	fs := NewFailoverState(h.maxAccountSwitches, false)
+
+	for {
+		selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionHash, reqModel, fs.FailedAccountIDs, "")
+		if err != nil {
+			if len(fs.FailedAccountIDs) == 0 {
+				h.chatCompletionsErrorResponse(c, http.StatusServiceUnavailable, "api_error", "No available accounts: "+err.Error())
+				return
+			}
+			action := fs.HandleSelectionExhausted(c.Request.Context())
+			switch action {
+			case FailoverContinue:
+				continue
+			case FailoverCanceled:
+				return
+			default:
+				if fs.LastFailoverErr != nil {
+					h.handleCCFailoverExhausted(c, fs.LastFailoverErr, streamStarted)
+				} else {
+					h.chatCompletionsErrorResponse(c, http.StatusBadGateway, "server_error", "All available accounts exhausted")
+				}
+				return
+			}
+		}
+		account := selection.Account
+		setOpsSelectedAccount(c, account.ID, account.Platform)
+
+		// 4. Acquire account concurrency slot
+		accountReleaseFunc := selection.ReleaseFunc
+		if !selection.Acquired {
+			if selection.WaitPlan == nil {
+				h.chatCompletionsErrorResponse(c, http.StatusServiceUnavailable, "api_error", "No available accounts")
+				return
+			}
+			accountReleaseFunc, err = h.concurrencyHelper.AcquireAccountSlotWithWaitTimeout(
+				c,
+				account.ID,
+				selection.WaitPlan.MaxConcurrency,
+				selection.WaitPlan.Timeout,
+				reqStream,
+				&streamStarted,
+			)
+			if err != nil {
+				reqLog.Warn("gateway.cc.account_slot_acquire_failed", zap.Int64("account_id", account.ID), zap.Error(err))
+				h.handleConcurrencyError(c, err, "account", streamStarted)
+				return
+			}
+		}
+		accountReleaseFunc = wrapReleaseOnDone(c.Request.Context(), accountReleaseFunc)
+
+		// 5. Forward request
+		writerSizeBeforeForward := c.Writer.Size()
+		result, err := h.gatewayService.ForwardAsChatCompletions(c.Request.Context(), c, account, body, parsedReq)
+
+		if accountReleaseFunc != nil {
+			accountReleaseFunc()
+		}
+
+		if err != nil {
+			var failoverErr *service.UpstreamFailoverError
+			if errors.As(err, &failoverErr) {
+				if c.Writer.Size() != writerSizeBeforeForward {
+					h.handleCCFailoverExhausted(c, failoverErr, true)
+					return
+				}
+				action := fs.HandleFailoverError(c.Request.Context(), h.gatewayService, account.ID, account.Platform, failoverErr)
+				switch action {
+				case FailoverContinue:
+					continue
+				case FailoverExhausted:
+					h.handleCCFailoverExhausted(c, fs.LastFailoverErr, streamStarted)
+					return
+				case FailoverCanceled:
+					return
+				}
+			}
+			h.ensureForwardErrorResponse(c, streamStarted)
+			reqLog.Error("gateway.cc.forward_failed",
+				zap.Int64("account_id", account.ID),
+				zap.Error(err),
+			)
+			return
+		}
+
+		// 6. Record usage
+		userAgent := c.GetHeader("User-Agent")
+		clientIP := ip.GetClientIP(c)
+		requestPayloadHash := service.HashUsageRequestPayload(body)
+		inboundEndpoint := GetInboundEndpoint(c)
+		upstreamEndpoint := GetUpstreamEndpoint(c, account.Platform)
+
+		h.submitUsageRecordTask(func(ctx context.Context) {
+			if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{
+				Result:             result,
+				APIKey:             apiKey,
+				User:               apiKey.User,
+				Account:            account,
+				Subscription:       subscription,
+				InboundEndpoint:    inboundEndpoint,
+				UpstreamEndpoint:   upstreamEndpoint,
+				UserAgent:          userAgent,
+				IPAddress:          clientIP,
+				RequestPayloadHash: requestPayloadHash,
+				APIKeyService:      h.apiKeyService,
+			}); err != nil {
+				reqLog.Error("gateway.cc.record_usage_failed",
+					zap.Int64("account_id", account.ID),
+					zap.Error(err),
+				)
+			}
+		})
+		return
+	}
+}
+
+// chatCompletionsErrorResponse writes an error in OpenAI Chat Completions format.
+func (h *GatewayHandler) chatCompletionsErrorResponse(c *gin.Context, status int, errType, message string) {
+	c.JSON(status, gin.H{
+		"error": gin.H{
+			"type":    errType,
+			"message": message,
+		},
+	})
+}
+
+// handleCCFailoverExhausted writes a failover-exhausted error in CC format.
+func (h *GatewayHandler) handleCCFailoverExhausted(c *gin.Context, lastErr *service.UpstreamFailoverError, streamStarted bool) {
+	if streamStarted {
+		return
+	}
+	statusCode := http.StatusBadGateway
+	if lastErr != nil && lastErr.StatusCode > 0 {
+		statusCode = lastErr.StatusCode
+	}
+	h.chatCompletionsErrorResponse(c, statusCode, "server_error", "All available accounts exhausted")
+}

backend/internal/handler/gateway_handler_responses.go

@@ -0,0 +1,295 @@
+package handler
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"time"
+
+	pkghttputil "github.com/Wei-Shaw/sub2api/internal/pkg/httputil"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/ip"
+	middleware2 "github.com/Wei-Shaw/sub2api/internal/server/middleware"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
+	"go.uber.org/zap"
+)
+
+// Responses handles OpenAI Responses API endpoint for Anthropic platform groups.
+// POST /v1/responses
+// This converts Responses API requests to Anthropic format, forwards to Anthropic
+// upstream, and converts responses back to Responses format.
+func (h *GatewayHandler) Responses(c *gin.Context) {
+	streamStarted := false
+
+	requestStart := time.Now()
+
+	apiKey, ok := middleware2.GetAPIKeyFromContext(c)
+	if !ok {
+		h.responsesErrorResponse(c, http.StatusUnauthorized, "authentication_error", "Invalid API key")
+		return
+	}
+
+	subject, ok := middleware2.GetAuthSubjectFromContext(c)
+	if !ok {
+		h.responsesErrorResponse(c, http.StatusInternalServerError, "api_error", "User context not found")
+		return
+	}
+	reqLog := requestLogger(
+		c,
+		"handler.gateway.responses",
+		zap.Int64("user_id", subject.UserID),
+		zap.Int64("api_key_id", apiKey.ID),
+		zap.Any("group_id", apiKey.GroupID),
+	)
+
+	// Read request body
+	body, err := pkghttputil.ReadRequestBodyWithPrealloc(c.Request)
+	if err != nil {
+		if maxErr, ok := extractMaxBytesError(err); ok {
+			h.responsesErrorResponse(c, http.StatusRequestEntityTooLarge, "invalid_request_error", buildBodyTooLargeMessage(maxErr.Limit))
+			return
+		}
+		h.responsesErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to read request body")
+		return
+	}
+
+	if len(body) == 0 {
+		h.responsesErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Request body is empty")
+		return
+	}
+
+	setOpsRequestContext(c, "", false, body)
+
+	// Validate JSON
+	if !gjson.ValidBytes(body) {
+		h.responsesErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "Failed to parse request body")
+		return
+	}
+
+	// Extract model and stream using gjson (like OpenAI handler)
+	modelResult := gjson.GetBytes(body, "model")
+	if !modelResult.Exists() || modelResult.Type != gjson.String || modelResult.String() == "" {
+		h.responsesErrorResponse(c, http.StatusBadRequest, "invalid_request_error", "model is required")
+		return
+	}
+	reqModel := modelResult.String()
+	reqStream := gjson.GetBytes(body, "stream").Bool()
+	reqLog = reqLog.With(zap.String("model", reqModel), zap.Bool("stream", reqStream))
+
+	setOpsRequestContext(c, reqModel, reqStream, body)
+	setOpsEndpointContext(c, "", int16(service.RequestTypeFromLegacy(reqStream, false)))
+
+	// Claude Code only restriction:
+	// /v1/responses is never a Claude Code endpoint.
+	// When claude_code_only is enabled, this endpoint is rejected.
+	// The existing service-layer checkClaudeCodeRestriction handles degradation
+	// to fallback groups when the Forward path calls SelectAccountForModelWithExclusions.
+	// Here we just reject at handler level since /v1/responses clients can't be Claude Code.
+	if apiKey.Group != nil && apiKey.Group.ClaudeCodeOnly {
+		h.responsesErrorResponse(c, http.StatusForbidden, "permission_error",
+			"This group is restricted to Claude Code clients (/v1/messages only)")
+		return
+	}
+
+	// Error passthrough binding
+	if h.errorPassthroughService != nil {
+		service.BindErrorPassthroughService(c, h.errorPassthroughService)
+	}
+
+	subscription, _ := middleware2.GetSubscriptionFromContext(c)
+
+	service.SetOpsLatencyMs(c, service.OpsAuthLatencyMsKey, time.Since(requestStart).Milliseconds())
+
+	// 1. Acquire user concurrency slot
+	maxWait := service.CalculateMaxWait(subject.Concurrency)
+	canWait, err := h.concurrencyHelper.IncrementWaitCount(c.Request.Context(), subject.UserID, maxWait)
+	waitCounted := false
+	if err != nil {
+		reqLog.Warn("gateway.responses.user_wait_counter_increment_failed", zap.Error(err))
+	} else if !canWait {
+		h.responsesErrorResponse(c, http.StatusTooManyRequests, "rate_limit_error", "Too many pending requests, please retry later")
+		return
+	}
+	if err == nil && canWait {
+		waitCounted = true
+	}
+	defer func() {
+		if waitCounted {
+			h.concurrencyHelper.DecrementWaitCount(c.Request.Context(), subject.UserID)
+		}
+	}()
+
+	userReleaseFunc, err := h.concurrencyHelper.AcquireUserSlotWithWait(c, subject.UserID, subject.Concurrency, reqStream, &streamStarted)
+	if err != nil {
+		reqLog.Warn("gateway.responses.user_slot_acquire_failed", zap.Error(err))
+		h.handleConcurrencyError(c, err, "user", streamStarted)
+		return
+	}
+	if waitCounted {
+		h.concurrencyHelper.DecrementWaitCount(c.Request.Context(), subject.UserID)
+		waitCounted = false
+	}
+	userReleaseFunc = wrapReleaseOnDone(c.Request.Context(), userReleaseFunc)
+	if userReleaseFunc != nil {
+		defer userReleaseFunc()
+	}
+
+	// 2. Re-check billing
+	if err := h.billingCacheService.CheckBillingEligibility(c.Request.Context(), apiKey.User, apiKey, apiKey.Group, subscription); err != nil {
+		reqLog.Info("gateway.responses.billing_check_failed", zap.Error(err))
+		status, code, message := billingErrorDetails(err)
+		h.responsesErrorResponse(c, status, code, message)
+		return
+	}
+
+	// Parse request for session hash
+	parsedReq, _ := service.ParseGatewayRequest(body, "responses")
+	if parsedReq == nil {
+		parsedReq = &service.ParsedRequest{Model: reqModel, Stream: reqStream, Body: body}
+	}
+	parsedReq.SessionContext = &service.SessionContext{
+		ClientIP:  ip.GetClientIP(c),
+		UserAgent: c.GetHeader("User-Agent"),
+		APIKeyID:  apiKey.ID,
+	}
+	sessionHash := h.gatewayService.GenerateSessionHash(parsedReq)
+
+	// 3. Account selection + failover loop
+	fs := NewFailoverState(h.maxAccountSwitches, false)
+
+	for {
+		selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionHash, reqModel, fs.FailedAccountIDs, "")
+		if err != nil {
+			if len(fs.FailedAccountIDs) == 0 {
+				h.responsesErrorResponse(c, http.StatusServiceUnavailable, "api_error", "No available accounts: "+err.Error())
+				return
+			}
+			action := fs.HandleSelectionExhausted(c.Request.Context())
+			switch action {
+			case FailoverContinue:
+				continue
+			case FailoverCanceled:
+				return
+			default:
+				if fs.LastFailoverErr != nil {
+					h.handleResponsesFailoverExhausted(c, fs.LastFailoverErr, streamStarted)
+				} else {
+					h.responsesErrorResponse(c, http.StatusBadGateway, "server_error", "All available accounts exhausted")
+				}
+				return
+			}
+		}
+		account := selection.Account
+		setOpsSelectedAccount(c, account.ID, account.Platform)
+
+		// 4. Acquire account concurrency slot
+		accountReleaseFunc := selection.ReleaseFunc
+		if !selection.Acquired {
+			if selection.WaitPlan == nil {
+				h.responsesErrorResponse(c, http.StatusServiceUnavailable, "api_error", "No available accounts")
+				return
+			}
+			accountReleaseFunc, err = h.concurrencyHelper.AcquireAccountSlotWithWaitTimeout(
+				c,
+				account.ID,
+				selection.WaitPlan.MaxConcurrency,
+				selection.WaitPlan.Timeout,
+				reqStream,
+				&streamStarted,
+			)
+			if err != nil {
+				reqLog.Warn("gateway.responses.account_slot_acquire_failed", zap.Int64("account_id", account.ID), zap.Error(err))
+				h.handleConcurrencyError(c, err, "account", streamStarted)
+				return
+			}
+		}
+		accountReleaseFunc = wrapReleaseOnDone(c.Request.Context(), accountReleaseFunc)
+
+		// 5. Forward request
+		writerSizeBeforeForward := c.Writer.Size()
+		result, err := h.gatewayService.ForwardAsResponses(c.Request.Context(), c, account, body, parsedReq)
+
+		if accountReleaseFunc != nil {
+			accountReleaseFunc()
+		}
+
+		if err != nil {
+			var failoverErr *service.UpstreamFailoverError
+			if errors.As(err, &failoverErr) {
+				// Can't failover if streaming content already sent
+				if c.Writer.Size() != writerSizeBeforeForward {
+					h.handleResponsesFailoverExhausted(c, failoverErr, true)
+					return
+				}
+				action := fs.HandleFailoverError(c.Request.Context(), h.gatewayService, account.ID, account.Platform, failoverErr)
+				switch action {
+				case FailoverContinue:
+					continue
+				case FailoverExhausted:
+					h.handleResponsesFailoverExhausted(c, fs.LastFailoverErr, streamStarted)
+					return
+				case FailoverCanceled:
+					return
+				}
+			}
+			h.ensureForwardErrorResponse(c, streamStarted)
+			reqLog.Error("gateway.responses.forward_failed",
+				zap.Int64("account_id", account.ID),
+				zap.Error(err),
+			)
+			return
+		}
+
+		// 6. Record usage
+		userAgent := c.GetHeader("User-Agent")
+		clientIP := ip.GetClientIP(c)
+		requestPayloadHash := service.HashUsageRequestPayload(body)
+		inboundEndpoint := GetInboundEndpoint(c)
+		upstreamEndpoint := GetUpstreamEndpoint(c, account.Platform)
+
+		h.submitUsageRecordTask(func(ctx context.Context) {
+			if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{
+				Result:             result,
+				APIKey:             apiKey,
+				User:               apiKey.User,
+				Account:            account,
+				Subscription:       subscription,
+				InboundEndpoint:    inboundEndpoint,
+				UpstreamEndpoint:   upstreamEndpoint,
+				UserAgent:          userAgent,
+				IPAddress:          clientIP,
+				RequestPayloadHash: requestPayloadHash,
+				APIKeyService:      h.apiKeyService,
+			}); err != nil {
+				reqLog.Error("gateway.responses.record_usage_failed",
+					zap.Int64("account_id", account.ID),
+					zap.Error(err),
+				)
+			}
+		})
+		return
+	}
+}
+
+// responsesErrorResponse writes an error in OpenAI Responses API format.
+func (h *GatewayHandler) responsesErrorResponse(c *gin.Context, status int, code, message string) {
+	c.JSON(status, gin.H{
+		"error": gin.H{
+			"code":    code,
+			"message": message,
+		},
+	})
+}
+
+// handleResponsesFailoverExhausted writes a failover-exhausted error in Responses format.
+func (h *GatewayHandler) handleResponsesFailoverExhausted(c *gin.Context, lastErr *service.UpstreamFailoverError, streamStarted bool) {
+	if streamStarted {
+		return // Can't write error after stream started
+	}
+	statusCode := http.StatusBadGateway
+	if lastErr != nil && lastErr.StatusCode > 0 {
+		statusCode = lastErr.StatusCode
+	}
+	h.responsesErrorResponse(c, statusCode, "server_error", "All available accounts exhausted")
+}

backend/internal/handler/gemini_v1beta_handler.go

@@ -182,6 +182,7 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
 	}
 
 	setOpsRequestContext(c, modelName, stream, body)
+	setOpsEndpointContext(c, "", int16(service.RequestTypeFromLegacy(stream, false)))
 
 	// Get subscription (may be nil)
 	subscription, _ := middleware.GetSubscriptionFromContext(c)

backend/internal/handler/openai_chat_completions.go

@@ -77,6 +77,7 @@ func (h *OpenAIGatewayHandler) ChatCompletions(c *gin.Context) {
 	reqLog = reqLog.With(zap.String("model", reqModel), zap.Bool("stream", reqStream))
 
 	setOpsRequestContext(c, reqModel, reqStream, body)
+	setOpsEndpointContext(c, "", int16(service.RequestTypeFromLegacy(reqStream, false)))
 
 	if h.errorPassthroughService != nil {
 		service.BindErrorPassthroughService(c, h.errorPassthroughService)

backend/internal/handler/openai_gateway_handler.go

@@ -183,6 +183,7 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 	}
 
 	setOpsRequestContext(c, reqModel, reqStream, body)
+	setOpsEndpointContext(c, "", int16(service.RequestTypeFromLegacy(reqStream, false)))
 
 	// 提前校验 function_call_output 是否具备可关联上下文，避免上游 400。
 	if !h.validateFunctionCallOutputRequest(c, body, reqLog) {
@@ -545,6 +546,7 @@ func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
 	reqLog = reqLog.With(zap.String("model", reqModel), zap.Bool("stream", reqStream))
 
 	setOpsRequestContext(c, reqModel, reqStream, body)
+	setOpsEndpointContext(c, "", int16(service.RequestTypeFromLegacy(reqStream, false)))
 
 	// 绑定错误透传服务，允许 service 层在非 failover 错误场景复用规则。
 	if h.errorPassthroughService != nil {
@@ -1096,6 +1098,7 @@ func (h *OpenAIGatewayHandler) ResponsesWebSocket(c *gin.Context) {
 		zap.String("previous_response_id_kind", previousResponseIDKind),
 	)
 	setOpsRequestContext(c, reqModel, true, firstMessage)
+	setOpsEndpointContext(c, "", int16(service.RequestTypeWSV2))
 
 	var currentUserRelease func()
 	var currentAccountRelease func()

backend/internal/handler/ops_error_logger.go

@@ -27,6 +27,9 @@ const (
 	opsRequestBodyKey = "ops_request_body"
 	opsAccountIDKey   = "ops_account_id"
 
+	opsUpstreamModelKey = "ops_upstream_model"
+	opsRequestTypeKey   = "ops_request_type"
+
 	// 错误过滤匹配常量 — shouldSkipOpsErrorLog 和错误分类共用
 	opsErrContextCanceled            = "context canceled"
 	opsErrNoAvailableAccounts        = "no available accounts"
@@ -345,6 +348,18 @@ func setOpsRequestContext(c *gin.Context, model string, stream bool, requestBody
 	}
 }
 
+// setOpsEndpointContext stores upstream model and request type for ops error logging.
+// Called by handlers after model mapping and request type determination.
+func setOpsEndpointContext(c *gin.Context, upstreamModel string, requestType int16) {
+	if c == nil {
+		return
+	}
+	if upstreamModel = strings.TrimSpace(upstreamModel); upstreamModel != "" {
+		c.Set(opsUpstreamModelKey, upstreamModel)
+	}
+	c.Set(opsRequestTypeKey, requestType)
+}
+
 func attachOpsRequestBodyToEntry(c *gin.Context, entry *service.OpsInsertErrorLogInput) {
 	if c == nil || entry == nil {
 		return
@@ -628,7 +643,30 @@ func OpsErrorLoggerMiddleware(ops *service.OpsService) gin.HandlerFunc {
 					}
 					return ""
 				}(),
-				Stream:    stream,
+				Stream:           stream,
+				InboundEndpoint:  GetInboundEndpoint(c),
+				UpstreamEndpoint: GetUpstreamEndpoint(c, platform),
+				RequestedModel:   modelName,
+				UpstreamModel: func() string {
+					if v, ok := c.Get(opsUpstreamModelKey); ok {
+						if s, ok := v.(string); ok {
+							return strings.TrimSpace(s)
+						}
+					}
+					return ""
+				}(),
+				RequestType: func() *int16 {
+					if v, ok := c.Get(opsRequestTypeKey); ok {
+						switch t := v.(type) {
+						case int16:
+							return &t
+						case int:
+							v16 := int16(t)
+							return &v16
+						}
+					}
+					return nil
+				}(),
 				UserAgent: c.GetHeader("User-Agent"),
 
 				ErrorPhase: "upstream",
@@ -756,7 +794,30 @@ func OpsErrorLoggerMiddleware(ops *service.OpsService) gin.HandlerFunc {
 				}
 				return ""
 			}(),
-			Stream:    stream,
+			Stream:           stream,
+			InboundEndpoint:  GetInboundEndpoint(c),
+			UpstreamEndpoint: GetUpstreamEndpoint(c, platform),
+			RequestedModel:   modelName,
+			UpstreamModel: func() string {
+				if v, ok := c.Get(opsUpstreamModelKey); ok {
+					if s, ok := v.(string); ok {
+						return strings.TrimSpace(s)
+					}
+				}
+				return ""
+			}(),
+			RequestType: func() *int16 {
+				if v, ok := c.Get(opsRequestTypeKey); ok {
+					switch t := v.(type) {
+					case int16:
+						return &t
+					case int:
+						v16 := int16(t)
+						return &v16
+					}
+				}
+				return nil
+			}(),
 			UserAgent: c.GetHeader("User-Agent"),
 
 			ErrorPhase:        phase,

backend/internal/handler/ops_error_logger_test.go

@@ -274,3 +274,48 @@ func TestNormalizeOpsErrorType(t *testing.T) {
 		})
 	}
 }
+
+func TestSetOpsEndpointContext_SetsContextKeys(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", nil)
+
+	setOpsEndpointContext(c, "claude-3-5-sonnet-20241022", int16(2)) // stream
+
+	v, ok := c.Get(opsUpstreamModelKey)
+	require.True(t, ok)
+	vStr, ok := v.(string)
+	require.True(t, ok)
+	require.Equal(t, "claude-3-5-sonnet-20241022", vStr)
+
+	rt, ok := c.Get(opsRequestTypeKey)
+	require.True(t, ok)
+	rtVal, ok := rt.(int16)
+	require.True(t, ok)
+	require.Equal(t, int16(2), rtVal)
+}
+
+func TestSetOpsEndpointContext_EmptyModelNotStored(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", nil)
+
+	setOpsEndpointContext(c, "", int16(1))
+
+	_, ok := c.Get(opsUpstreamModelKey)
+	require.False(t, ok, "empty upstream model should not be stored")
+
+	rt, ok := c.Get(opsRequestTypeKey)
+	require.True(t, ok)
+	rtVal, ok := rt.(int16)
+	require.True(t, ok)
+	require.Equal(t, int16(1), rtVal)
+}
+
+func TestSetOpsEndpointContext_NilContext(t *testing.T) {
+	require.NotPanics(t, func() {
+		setOpsEndpointContext(nil, "model", int16(1))
+	})
+}

backend/internal/handler/sora_gateway_handler.go

@@ -159,6 +159,7 @@ func (h *SoraGatewayHandler) ChatCompletions(c *gin.Context) {
 	}
 
 	setOpsRequestContext(c, reqModel, clientStream, body)
+	setOpsEndpointContext(c, "", int16(service.RequestTypeFromLegacy(clientStream, false)))
 
 	platform := ""
 	if forced, ok := middleware2.GetForcePlatformFromContext(c); ok {