refactor: audio relay

relay/relay-audio.go

@@ -1,13 +1,10 @@
 package relay
 
 import (
-	"bytes"
-	"context"
 	"encoding/json"
 	"errors"
 	"fmt"
 	"github.com/gin-gonic/gin"
-	"io"
 	"net/http"
 	"one-api/common"
 	"one-api/constant"
@@ -16,69 +13,71 @@ import (
 	relaycommon "one-api/relay/common"
 	relayconstant "one-api/relay/constant"
 	"one-api/service"
-	"strings"
-	"time"
 )
 
-func AudioHelper(c *gin.Context, relayMode int) *dto.OpenAIErrorWithStatusCode {
-	tokenId := c.GetInt("token_id")
-	channelType := c.GetInt("channel")
-	channelId := c.GetInt("channel_id")
-	userId := c.GetInt("id")
-	group := c.GetString("group")
-	startTime := time.Now()
-
-	var audioRequest dto.TextToSpeechRequest
-	if !strings.HasPrefix(c.Request.URL.Path, "/v1/audio/transcriptions") {
-		err := common.UnmarshalBodyReusable(c, &audioRequest)
-		if err != nil {
-			return service.OpenAIErrorWrapper(err, "bind_request_body_failed", http.StatusBadRequest)
-		}
-	} else {
-		audioRequest = dto.TextToSpeechRequest{
-			Model: "whisper-1",
-		}
+func getAndValidAudioRequest(c *gin.Context, info *relaycommon.RelayInfo) (*dto.AudioRequest, error) {
+	audioRequest := &dto.AudioRequest{}
+	err := common.UnmarshalBodyReusable(c, audioRequest)
+	if err != nil {
+		return nil, err
 	}
-	//err := common.UnmarshalBodyReusable(c, &audioRequest)
-
-	// request validation
-	if audioRequest.Model == "" {
-		return service.OpenAIErrorWrapper(errors.New("model is required"), "required_field_missing", http.StatusBadRequest)
-	}
-
-	if strings.HasPrefix(audioRequest.Model, "tts-1") {
-		if audioRequest.Voice == "" {
-			return service.OpenAIErrorWrapper(errors.New("voice is required"), "required_field_missing", http.StatusBadRequest)
+	switch info.RelayMode {
+	case relayconstant.RelayModeAudioSpeech:
+		if audioRequest.Model == "" {
+			return nil, errors.New("model is required")
 		}
-	}
-	var err error
-	promptTokens := 0
-	preConsumedTokens := common.PreConsumedQuota
-	if strings.HasPrefix(audioRequest.Model, "tts-1") {
 		if constant.ShouldCheckPromptSensitive() {
-			err = service.CheckSensitiveInput(audioRequest.Input)
+			err := service.CheckSensitiveInput(audioRequest.Input)
 			if err != nil {
-				return service.OpenAIErrorWrapper(err, "sensitive_words_detected", http.StatusBadRequest)
+				return nil, err
 			}
 		}
+	default:
+		if audioRequest.Model == "" {
+			audioRequest.Model = c.PostForm("model")
+		}
+		if audioRequest.Model == "" {
+			return nil, errors.New("model is required")
+		}
+		if audioRequest.ResponseFormat == "" {
+			audioRequest.ResponseFormat = "json"
+		}
+	}
+	return audioRequest, nil
+}
+
+func AudioHelper(c *gin.Context, relayMode int) *dto.OpenAIErrorWithStatusCode {
+	relayInfo := relaycommon.GenRelayInfo(c)
+	audioRequest, err := getAndValidAudioRequest(c, relayInfo)
+
+	if err != nil {
+		common.LogError(c, fmt.Sprintf("getAndValidAudioRequest failed: %s", err.Error()))
+		return service.OpenAIErrorWrapper(err, "invalid_audio_request", http.StatusBadRequest)
+	}
+
+	promptTokens := 0
+	preConsumedTokens := common.PreConsumedQuota
+	if relayInfo.RelayMode == relayconstant.RelayModeAudioSpeech {
 		promptTokens, err = service.CountAudioToken(audioRequest.Input, audioRequest.Model)
 		if err != nil {
 			return service.OpenAIErrorWrapper(err, "count_audio_token_failed", http.StatusInternalServerError)
 		}
 		preConsumedTokens = promptTokens
+		relayInfo.PromptTokens = promptTokens
 	}
+
 	modelRatio := common.GetModelRatio(audioRequest.Model)
-	groupRatio := common.GetGroupRatio(group)
+	groupRatio := common.GetGroupRatio(relayInfo.Group)
 	ratio := modelRatio * groupRatio
 	preConsumedQuota := int(float64(preConsumedTokens) * ratio)
-	userQuota, err := model.CacheGetUserQuota(userId)
+	userQuota, err := model.CacheGetUserQuota(relayInfo.UserId)
 	if err != nil {
 		return service.OpenAIErrorWrapperLocal(err, "get_user_quota_failed", http.StatusInternalServerError)
 	}
 	if userQuota-preConsumedQuota < 0 {
 		return service.OpenAIErrorWrapperLocal(errors.New("user quota is not enough"), "insufficient_user_quota", http.StatusForbidden)
 	}
-	err = model.CacheDecreaseUserQuota(userId, preConsumedQuota)
+	err = model.CacheDecreaseUserQuota(relayInfo.UserId, preConsumedQuota)
 	if err != nil {
 		return service.OpenAIErrorWrapperLocal(err, "decrease_user_quota_failed", http.StatusInternalServerError)
 	}
@@ -88,28 +87,12 @@ func AudioHelper(c *gin.Context, relayMode int) *dto.OpenAIErrorWithStatusCode {
 		preConsumedQuota = 0
 	}
 	if preConsumedQuota > 0 {
-		userQuota, err = model.PreConsumeTokenQuota(tokenId, preConsumedQuota)
+		userQuota, err = model.PreConsumeTokenQuota(relayInfo.TokenId, preConsumedQuota)
 		if err != nil {
 			return service.OpenAIErrorWrapperLocal(err, "pre_consume_token_quota_failed", http.StatusForbidden)
 		}
 	}
 
-	succeed := false
-	defer func() {
-		if succeed {
-			return
-		}
-		if preConsumedQuota > 0 {
-			// we need to roll back the pre-consumed quota
-			defer func() {
-				go func() {
-					// negative means add quota back for token & user
-					returnPreConsumedQuota(c, tokenId, userQuota, preConsumedQuota)
-				}()
-			}()
-		}
-	}()
-
 	// map model name
 	modelMapping := c.GetString("model_mapping")
 	if modelMapping != "" {
@@ -123,132 +106,42 @@ func AudioHelper(c *gin.Context, relayMode int) *dto.OpenAIErrorWithStatusCode {
 		}
 	}
 
-	baseURL := common.ChannelBaseURLs[channelType]
-	requestURL := c.Request.URL.String()
-	if c.GetString("base_url") != "" {
-		baseURL = c.GetString("base_url")
+	adaptor := GetAdaptor(relayInfo.ApiType)
+	if adaptor == nil {
+		return service.OpenAIErrorWrapperLocal(fmt.Errorf("invalid api type: %d", relayInfo.ApiType), "invalid_api_type", http.StatusBadRequest)
 	}
+	adaptor.Init(relayInfo)
 
-	fullRequestURL := relaycommon.GetFullRequestURL(baseURL, requestURL, channelType)
-	if relayMode == relayconstant.RelayModeAudioTranscription && channelType == common.ChannelTypeAzure {
-		// https://learn.microsoft.com/en-us/azure/ai-services/openai/whisper-quickstart?tabs=command-line#rest-api
-		apiVersion := relaycommon.GetAPIVersion(c)
-		fullRequestURL = fmt.Sprintf("%s/openai/deployments/%s/audio/transcriptions?api-version=%s", baseURL, audioRequest.Model, apiVersion)
-	}
-
-	requestBody := c.Request.Body
-
-	req, err := http.NewRequest(c.Request.Method, fullRequestURL, requestBody)
+	ioReader, err := adaptor.ConvertAudioRequest(c, relayInfo, *audioRequest)
 	if err != nil {
-		return service.OpenAIErrorWrapper(err, "new_request_failed", http.StatusInternalServerError)
+		return service.OpenAIErrorWrapperLocal(err, "convert_request_failed", http.StatusInternalServerError)
 	}
 
-	if relayMode == relayconstant.RelayModeAudioTranscription && channelType == common.ChannelTypeAzure {
-		// https://learn.microsoft.com/en-us/azure/ai-services/openai/whisper-quickstart?tabs=command-line#rest-api
-		apiKey := c.Request.Header.Get("Authorization")
-		apiKey = strings.TrimPrefix(apiKey, "Bearer ")
-		req.Header.Set("api-key", apiKey)
-		req.ContentLength = c.Request.ContentLength
-	} else {
-		req.Header.Set("Authorization", c.Request.Header.Get("Authorization"))
-	}
-
-	req.Header.Set("Content-Type", c.Request.Header.Get("Content-Type"))
-	req.Header.Set("Accept", c.Request.Header.Get("Accept"))
-
-	resp, err := service.GetHttpClient().Do(req)
+	resp, err := adaptor.DoRequest(c, relayInfo, ioReader)
 	if err != nil {
 		return service.OpenAIErrorWrapper(err, "do_request_failed", http.StatusInternalServerError)
 	}
 
-	err = req.Body.Close()
-	if err != nil {
-		return service.OpenAIErrorWrapper(err, "close_request_body_failed", http.StatusInternalServerError)
-	}
-	err = c.Request.Body.Close()
-	if err != nil {
-		return service.OpenAIErrorWrapper(err, "close_request_body_failed", http.StatusInternalServerError)
-	}
-
-	if resp.StatusCode != http.StatusOK {
-		return relaycommon.RelayErrorHandler(resp)
-	}
-	succeed = true
-
-	var audioResponse dto.AudioResponse
-
-	defer func(ctx context.Context) {
-		go func() {
-			useTimeSeconds := time.Now().Unix() - startTime.Unix()
-			quota := 0
-			if strings.HasPrefix(audioRequest.Model, "tts-1") {
-				quota = promptTokens
-			} else {
-				quota, err = service.CountAudioToken(audioResponse.Text, audioRequest.Model)
-			}
-			quota = int(float64(quota) * ratio)
-			if ratio != 0 && quota <= 0 {
-				quota = 1
-			}
-			quotaDelta := quota - preConsumedQuota
-			err := model.PostConsumeTokenQuota(tokenId, userQuota, quotaDelta, preConsumedQuota, true)
-			if err != nil {
-				common.SysError("error consuming token remain quota: " + err.Error())
-			}
-			err = model.CacheUpdateUserQuota(userId)
-			if err != nil {
-				common.SysError("error update user quota cache: " + err.Error())
-			}
-			if quota != 0 {
-				tokenName := c.GetString("token_name")
-				logContent := fmt.Sprintf("模型倍率 %.2f，分组倍率 %.2f", modelRatio, groupRatio)
-				other := make(map[string]interface{})
-				other["model_ratio"] = modelRatio
-				other["group_ratio"] = groupRatio
-				model.RecordConsumeLog(ctx, userId, channelId, promptTokens, 0, audioRequest.Model, tokenName, quota, logContent, tokenId, userQuota, int(useTimeSeconds), false, other)
-				model.UpdateUserUsedQuotaAndRequestCount(userId, quota)
-				channelId := c.GetInt("channel_id")
-				model.UpdateChannelUsedQuota(channelId, quota)
-			}
-		}()
-	}(c.Request.Context())
-
-	responseBody, err := io.ReadAll(resp.Body)
-
-	if err != nil {
-		return service.OpenAIErrorWrapper(err, "read_response_body_failed", http.StatusInternalServerError)
-	}
-	err = resp.Body.Close()
-	if err != nil {
-		return service.OpenAIErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError)
-	}
-	if strings.HasPrefix(audioRequest.Model, "tts-1") {
-
-	} else {
-		err = json.Unmarshal(responseBody, &audioResponse)
-		if err != nil {
-			return service.OpenAIErrorWrapper(err, "unmarshal_response_body_failed", http.StatusInternalServerError)
-		}
-		contains, words := service.SensitiveWordContains(audioResponse.Text)
-		if contains {
-			return service.OpenAIErrorWrapper(errors.New("response contains sensitive words: "+strings.Join(words, ", ")), "response_contains_sensitive_words", http.StatusBadRequest)
+	statusCodeMappingStr := c.GetString("status_code_mapping")
+	if resp != nil {
+		if resp.StatusCode != http.StatusOK {
+			returnPreConsumedQuota(c, relayInfo.TokenId, userQuota, preConsumedQuota)
+			openaiErr := service.RelayErrorHandler(resp)
+			// reset status code 重置状态码
+			service.ResetStatusCode(openaiErr, statusCodeMappingStr)
+			return openaiErr
 		}
 	}
 
-	resp.Body = io.NopCloser(bytes.NewBuffer(responseBody))
+	usage, openaiErr := adaptor.DoResponse(c, resp, relayInfo)
+	if openaiErr != nil {
+		returnPreConsumedQuota(c, relayInfo.TokenId, userQuota, preConsumedQuota)
+		// reset status code 重置状态码
+		service.ResetStatusCode(openaiErr, statusCodeMappingStr)
+		return openaiErr
+	}
 
-	for k, v := range resp.Header {
-		c.Writer.Header().Set(k, v[0])
-	}
-	c.Writer.WriteHeader(resp.StatusCode)
+	postConsumeQuota(c, relayInfo, audioRequest.Model, usage, ratio, preConsumedQuota, userQuota, modelRatio, groupRatio, 0, false)
 
-	_, err = io.Copy(c.Writer, resp.Body)
-	if err != nil {
-		return service.OpenAIErrorWrapper(err, "copy_response_body_failed", http.StatusInternalServerError)
-	}
-	err = resp.Body.Close()
-	if err != nil {
-		return service.OpenAIErrorWrapper(err, "close_response_body_failed", http.StatusInternalServerError)
-	}
 	return nil
 }