ai-gateway/controller/relay.go

158 lines
4.6 KiB
Go
Raw Normal View History

2023-04-23 10:24:11 +00:00
package controller
import (
"bytes"
"context"
2023-04-23 10:24:11 +00:00
"fmt"
"io"
"net/http"
"github.com/gin-gonic/gin"
"github.com/songquanpeng/one-api/common"
2024-01-28 11:38:58 +00:00
"github.com/songquanpeng/one-api/common/config"
feat: support aws bedrockruntime claude3 (#1328) * feat: support aws bedrockruntime claude3 closes #622, closes #749, closes #1300 * fix: convert to aws claude model id * fix: Update AWS adapter to handle stream completions and calculate usage metrics Based on the file summaries provided, here are the important bullet points for the commit message: - Add functionality to handle stream completion events from AWS in the relay/adaptor/aws/main.go file - Marshall AWS response to OpenAI format and calculate usage metrics in the same file - Implement a custom render function for streaming events in the same file - Improve error handling for JSON unmarshalling and marshalling errors in the same file * fix: Implement AWS handler with usage tracking and error handling - Implemented streaming response handling for AWS handler - Set response content type to text/event-stream - Added error handling for failed marshaling/unmarshaling - Updated return values to include `relaymodel.ErrorWithStatusCode` and `relaymodel.Usage` - Improved error handling and response formatting for AWS adaptor * fix: Refactor AWS Adapter for Improved Model Mapping and Error Handling * Refactor AWS adapter to improve model management - Replace hardcoded model list in `adapter.go` with a function to get models from `awsModelIDMap` - Update `GetModelList` function to return model list directly - Add `GetChannelName` function to get channel name from `Adaptor` object * Improve error handling and code organization in main.go - Replace switch statement with a map to map AWS model IDs to OpenAI model IDs - Return an error if the model is not found in the map - Use a single return statement instead of wrapping multiple return statements in the `awsModelID` function - Add a new error message for when the model is not found in the map in the `Handler` function * fix: bug fix * chore: change variable name & package * chore: change variable name * perf: update config related code --------- Co-authored-by: JustSong <songquanpeng@foxmail.com>
2024-04-19 16:40:47 +00:00
"github.com/songquanpeng/one-api/common/ctxkey"
2024-01-28 11:38:58 +00:00
"github.com/songquanpeng/one-api/common/helper"
"github.com/songquanpeng/one-api/common/logger"
"github.com/songquanpeng/one-api/middleware"
dbmodel "github.com/songquanpeng/one-api/model"
"github.com/songquanpeng/one-api/monitor"
2024-01-28 11:38:58 +00:00
"github.com/songquanpeng/one-api/relay/controller"
"github.com/songquanpeng/one-api/relay/model"
"github.com/songquanpeng/one-api/relay/relaymode"
2023-04-23 10:24:11 +00:00
)
2023-05-21 06:26:59 +00:00
// https://platform.openai.com/docs/api-reference/chat
func relayHelper(c *gin.Context, relayMode int) *model.ErrorWithStatusCode {
var err *model.ErrorWithStatusCode
2023-06-19 02:28:55 +00:00
switch relayMode {
case relaymode.ImagesGenerations:
2024-01-14 11:21:03 +00:00
err = controller.RelayImageHelper(c, relayMode)
case relaymode.AudioSpeech:
fallthrough
case relaymode.AudioTranslation:
fallthrough
case relaymode.AudioTranscription:
2024-01-14 11:21:03 +00:00
err = controller.RelayAudioHelper(c, relayMode)
case relaymode.Proxy:
err = controller.RelayProxyHelper(c, relayMode)
2023-06-19 02:28:55 +00:00
default:
err = controller.RelayTextHelper(c)
}
return err
}
func Relay(c *gin.Context) {
ctx := c.Request.Context()
relayMode := relaymode.GetByPath(c.Request.URL.Path)
if config.DebugEnabled {
requestBody, _ := common.GetRequestBody(c)
logger.Debugf(ctx, "request body: %s", string(requestBody))
}
2024-04-21 11:43:23 +00:00
channelId := c.GetInt(ctxkey.ChannelId)
userId := c.GetInt(ctxkey.Id)
bizErr := relayHelper(c, relayMode)
if bizErr == nil {
monitor.Emit(channelId, true)
return
}
lastFailedChannelId := channelId
2024-04-21 11:43:23 +00:00
channelName := c.GetString(ctxkey.ChannelName)
group := c.GetString(ctxkey.Group)
feat: support aws bedrockruntime claude3 (#1328) * feat: support aws bedrockruntime claude3 closes #622, closes #749, closes #1300 * fix: convert to aws claude model id * fix: Update AWS adapter to handle stream completions and calculate usage metrics Based on the file summaries provided, here are the important bullet points for the commit message: - Add functionality to handle stream completion events from AWS in the relay/adaptor/aws/main.go file - Marshall AWS response to OpenAI format and calculate usage metrics in the same file - Implement a custom render function for streaming events in the same file - Improve error handling for JSON unmarshalling and marshalling errors in the same file * fix: Implement AWS handler with usage tracking and error handling - Implemented streaming response handling for AWS handler - Set response content type to text/event-stream - Added error handling for failed marshaling/unmarshaling - Updated return values to include `relaymodel.ErrorWithStatusCode` and `relaymodel.Usage` - Improved error handling and response formatting for AWS adaptor * fix: Refactor AWS Adapter for Improved Model Mapping and Error Handling * Refactor AWS adapter to improve model management - Replace hardcoded model list in `adapter.go` with a function to get models from `awsModelIDMap` - Update `GetModelList` function to return model list directly - Add `GetChannelName` function to get channel name from `Adaptor` object * Improve error handling and code organization in main.go - Replace switch statement with a map to map AWS model IDs to OpenAI model IDs - Return an error if the model is not found in the map - Use a single return statement instead of wrapping multiple return statements in the `awsModelID` function - Add a new error message for when the model is not found in the map in the `Handler` function * fix: bug fix * chore: change variable name & package * chore: change variable name * perf: update config related code --------- Co-authored-by: JustSong <songquanpeng@foxmail.com>
2024-04-19 16:40:47 +00:00
originalModel := c.GetString(ctxkey.OriginalModel)
go processChannelRelayError(ctx, userId, channelId, channelName, bizErr)
2024-04-26 15:05:48 +00:00
requestId := c.GetString(helper.RequestIdKey)
retryTimes := config.RetryTimes
if !shouldRetry(c, bizErr.StatusCode) {
logger.Errorf(ctx, "relay error happen, status code is %d, won't retry in this case", bizErr.StatusCode)
retryTimes = 0
}
for i := retryTimes; i > 0; i-- {
channel, err := dbmodel.CacheGetRandomSatisfiedChannel(group, originalModel, i != retryTimes)
if err != nil {
feat: support aws bedrockruntime claude3 (#1328) * feat: support aws bedrockruntime claude3 closes #622, closes #749, closes #1300 * fix: convert to aws claude model id * fix: Update AWS adapter to handle stream completions and calculate usage metrics Based on the file summaries provided, here are the important bullet points for the commit message: - Add functionality to handle stream completion events from AWS in the relay/adaptor/aws/main.go file - Marshall AWS response to OpenAI format and calculate usage metrics in the same file - Implement a custom render function for streaming events in the same file - Improve error handling for JSON unmarshalling and marshalling errors in the same file * fix: Implement AWS handler with usage tracking and error handling - Implemented streaming response handling for AWS handler - Set response content type to text/event-stream - Added error handling for failed marshaling/unmarshaling - Updated return values to include `relaymodel.ErrorWithStatusCode` and `relaymodel.Usage` - Improved error handling and response formatting for AWS adaptor * fix: Refactor AWS Adapter for Improved Model Mapping and Error Handling * Refactor AWS adapter to improve model management - Replace hardcoded model list in `adapter.go` with a function to get models from `awsModelIDMap` - Update `GetModelList` function to return model list directly - Add `GetChannelName` function to get channel name from `Adaptor` object * Improve error handling and code organization in main.go - Replace switch statement with a map to map AWS model IDs to OpenAI model IDs - Return an error if the model is not found in the map - Use a single return statement instead of wrapping multiple return statements in the `awsModelID` function - Add a new error message for when the model is not found in the map in the `Handler` function * fix: bug fix * chore: change variable name & package * chore: change variable name * perf: update config related code --------- Co-authored-by: JustSong <songquanpeng@foxmail.com>
2024-04-19 16:40:47 +00:00
logger.Errorf(ctx, "CacheGetRandomSatisfiedChannel failed: %+v", err)
break
2023-07-15 11:06:51 +00:00
}
logger.Infof(ctx, "using channel #%d to retry (remain times %d)", channel.Id, i)
if channel.Id == lastFailedChannelId {
continue
}
middleware.SetupContextForSelectedChannel(c, channel, originalModel)
requestBody, err := common.GetRequestBody(c)
c.Request.Body = io.NopCloser(bytes.NewBuffer(requestBody))
bizErr = relayHelper(c, relayMode)
if bizErr == nil {
return
}
2024-04-21 11:43:23 +00:00
channelId := c.GetInt(ctxkey.ChannelId)
lastFailedChannelId = channelId
2024-04-21 11:43:23 +00:00
channelName := c.GetString(ctxkey.ChannelName)
// BUG: bizErr is in race condition
go processChannelRelayError(ctx, userId, channelId, channelName, bizErr)
}
if bizErr != nil {
if bizErr.StatusCode == http.StatusTooManyRequests {
bizErr.Error.Message = "当前分组上游负载已饱和,请稍后再试"
}
// BUG: bizErr is in race condition
bizErr.Error.Message = helper.MessageWithRequestId(bizErr.Error.Message, requestId)
c.JSON(bizErr.StatusCode, gin.H{
"error": bizErr.Error,
})
}
}
func shouldRetry(c *gin.Context, statusCode int) bool {
2024-04-21 11:43:23 +00:00
if _, ok := c.Get(ctxkey.SpecificChannelId); ok {
return false
}
if statusCode == http.StatusTooManyRequests {
return true
}
if statusCode/100 == 5 {
return true
}
if statusCode == http.StatusBadRequest {
return false
}
if statusCode/100 == 2 {
return false
}
return true
}
func processChannelRelayError(ctx context.Context, userId int, channelId int, channelName string, err *model.ErrorWithStatusCode) {
logger.Errorf(ctx, "relay error (channel id %d, user id: %d): %s", channelId, userId, err.Message)
// https://platform.openai.com/docs/guides/error-codes/api-errors
if monitor.ShouldDisableChannel(&err.Error, err.StatusCode) {
monitor.DisableChannel(channelId, channelName, err.Message)
} else {
monitor.Emit(channelId, false)
}
}
func RelayNotImplemented(c *gin.Context) {
err := model.Error{
Message: "API not implemented",
Type: "one_api_error",
Param: "",
Code: "api_not_implemented",
}
2023-06-23 14:59:44 +00:00
c.JSON(http.StatusNotImplemented, gin.H{
"error": err,
})
}
func RelayNotFound(c *gin.Context) {
err := model.Error{
2023-08-11 11:53:01 +00:00
Message: fmt.Sprintf("Invalid URL (%s %s)", c.Request.Method, c.Request.URL.Path),
Type: "invalid_request_error",
Param: "",
2023-08-11 11:53:01 +00:00
Code: "",
}
2023-06-23 14:59:44 +00:00
c.JSON(http.StatusNotFound, gin.H{
"error": err,
})
}