feat: sync deployed v1.0.0-rc.38-custom.67

middleware/auth.go

@@ -450,6 +450,11 @@ func TokenAuth() func(c *gin.Context) {
 			return
 		}
 
+		if err := model.LoadPersonalGroupRatios(c, userCache.Id); err != nil {
+			abortWithOpenAiMessage(c, http.StatusServiceUnavailable, "User pricing temporarily unavailable")
+			return
+		}
+		defer service.ReleasePoolProbes(c)
 		userCache.WriteContext(c)
 
 		userGroup := userCache.Group

middleware/distributor.go

@@ -27,13 +27,16 @@ import (
 )
 
 type ModelRequest struct {
-	Model string `json:"model"`
-	Group string `json:"group,omitempty"`
+	Model  string `json:"model"`
+	Group  string `json:"group,omitempty"`
+	Stream bool   `json:"stream"`
 }
 
 func Distribute() func(c *gin.Context) {
 	return func(c *gin.Context) {
 		var channel *model.Channel
+		var requestedStream bool
+		rejectionMessage := "No available channel for this request."
 		defer func() {
 			if c.Writer.Status() >= 400 {
 				service.RecordRequestPolicyTermination(c, types.NewErrorWithStatusCode(errors.New("request rejected"), types.ErrorCodeInvalidRequest, c.Writer.Status(), types.ErrOptionWithSkipRetry()))
@@ -42,7 +45,7 @@ func Distribute() func(c *gin.Context) {
 					other := model.NewLogOther()
 					other.SetPublic("status_code", c.Writer.Status())
 					service.AppendRelayLogAdminInfo(c, nil, other)
-					model.RecordErrorLog(c, c.GetInt("id"), 0, c.GetString("original_model"), c.GetString("token_name"), "No available channel for this request.", c.GetInt("token_id"), 0, false, "", other)
+					model.RecordErrorLog(c, c.GetInt("id"), 0, c.GetString("original_model"), c.GetString("token_name"), rejectionMessage, c.GetInt("token_id"), 0, requestedStream, "", other)
 				}
 			}
 		}()
@@ -57,6 +60,8 @@ func Distribute() func(c *gin.Context) {
 			abortWithOpenAiMessage(c, http.StatusBadRequest, i18n.T(c, i18n.MsgDistributorInvalidRequest, map[string]any{"Error": err.Error()}))
 			return
 		}
+		requestedStream = modelRequest.Stream || strings.Contains(c.Request.URL.Path, "streamGenerateContent") || strings.EqualFold(c.GetHeader("Upgrade"), "websocket")
+		c.Set("pool_request_stream", requestedStream)
 		_, pinned, _ := constraints.ResolvedPin()
 		if !pinned {
 			// Select a channel for the user
@@ -124,6 +129,17 @@ func Distribute() func(c *gin.Context) {
 				} else if selectErr.MessageID != "" {
 					message = i18n.T(c, selectErr.MessageID, selectErr.Params)
 				}
+				if service.RequestRoutingPolicy(c) != nil && selectErr.NoAvailableChannel {
+					if c.GetBool("pool_degradation_unavailable") {
+						message = "Group quality checks are temporarily unavailable. Please retry later."
+					} else if wait := c.GetInt("pool_retry_after"); wait > 0 {
+						message = fmt.Sprintf("No eligible group is currently available; failed groups are cooling down or being checked. Retry after %d seconds.", wait)
+						c.Header("Retry-After", fmt.Sprint(wait))
+					} else {
+						message = "No eligible group is available for this model under the current pool and quality settings."
+					}
+				}
+				rejectionMessage = message
 				abortWithOpenAiMessage(c, selectErr.StatusCode, message, selectErr.Code)
 				return
 			}
@@ -295,8 +311,9 @@ func getModelFromJSONBody(c *gin.Context) (*ModelRequest, error) {
 	c.Request.Body = io.NopCloser(storage)
 
 	return &ModelRequest{
-		Model: model,
-		Group: group,
+		Model:  model,
+		Group:  group,
+		Stream: gjson.GetBytes(requestBody, "stream").Bool(),
 	}, nil
 }