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

relay/responses_websocket.go

@@ -281,7 +281,8 @@ func (s *responsesWSSession) runCall(c *gin.Context, state *responsesWSCallState
 		}
 	} else {
 		retry := &service.RetryParam{Ctx: c, TokenGroup: common.GetContextKeyString(c, appconstant.ContextKeyUsingGroup), ModelName: modelName, RequestPath: c.Request.URL.Path, Retry: common.GetPointer(0)}
-		for ; retry.GetRetry() <= common.RetryTimes; retry.IncreaseRetry() {
+		maxRetries := max(common.RetryTimes, service.PoolRetryBudget(c, modelName))
+		for ; retry.GetRetry() <= maxRetries; retry.IncreaseRetry() {
 			var channel *appmodel.Channel
 			channel, apiErr = selectResponsesWSChannel(c, modelName, retry)
 			if apiErr != nil {
@@ -316,7 +317,15 @@ func (s *responsesWSSession) runCall(c *gin.Context, state *responsesWSCallState
 				apiErr = service.NormalizeViolationFeeError(types.NewError(dialErr, types.ErrorCodeDoRequestFailed))
 				service.ResetStatusCode(apiErr, c.GetString("status_code_mapping"))
 				info.LastError = apiErr
-				decision := service.DecideRelayRetry(c, apiErr, common.RetryTimes-retry.GetRetry())
+				remaining := common.RetryTimes - retry.GetRetry()
+				if service.IsPoolFailoverError(c, apiErr) {
+					remaining = maxRetries - retry.GetRetry()
+					service.RecordPoolRetryableFailure(c, modelName, info.UsingGroup)
+				}
+				decision := service.DecideRelayRetry(c, apiErr, remaining)
+				if service.IsPoolFailoverError(c, apiErr) && service.PoolRetryExhausted(c, modelName, info.UsingGroup) {
+					decision = service.PolicyDecision{Action: "stop", Reason: "pool_no_backup", Source: "routing"}
+				}
 				service.RecordPolicyFailure(c, channel.Id, apiErr, decision)
 				service.ProcessChannelError(c, *types.NewChannelError(channel.Id, channel.Type, channel.Name, channel.ChannelInfo.IsMultiKey, info.ApiKey, channel.GetAutoBan()), apiErr, info)
 				if decision.Action == "retry" {
@@ -571,6 +580,13 @@ func (s *responsesWSSession) restoreConnectionContext(c *gin.Context, model stri
 		}
 	} else {
 		group := common.GetContextKeyString(c, appconstant.ContextKeyUsingGroup)
+		if service.RequestRoutingPolicy(c) != nil {
+			if !slices.ContainsFunc(service.ResolveRequestPool(c, model), func(entry appmodel.TokenGroupEntry) bool { return entry.Group == s.lockedGroup }) {
+				return types.NewErrorWithStatusCode(errors.New("the connection group is no longer allowed"), types.ErrorCodeAccessDenied, http.StatusForbidden, types.ErrOptionWithSkipRetry())
+			}
+			group = s.lockedGroup
+			common.SetContextKey(c, appconstant.ContextKeyAutoGroup, group)
+		}
 		if group == "auto" {
 			if !slices.Contains(service.GetRequestAutoGroups(c, common.GetContextKeyString(c, appconstant.ContextKeyUserGroup)), s.lockedGroup) {
 				return types.NewErrorWithStatusCode(errors.New("the connection group is no longer allowed"), types.ErrorCodeAccessDenied, http.StatusForbidden, types.ErrOptionWithSkipRetry())