feat: 支持 API Key 上游池模式同账号重试次数配置与自定义错误策略

This commit is contained in:
kyx236
2026-03-08 13:57:23 +08:00
parent 03bf348530
commit e643fc382c
13 changed files with 558 additions and 40 deletions

View File

@@ -20,6 +20,7 @@ import (
coderws "github.com/coder/websocket"
"github.com/gin-gonic/gin"
"github.com/google/uuid"
"github.com/tidwall/gjson"
"go.uber.org/zap"
)
@@ -212,6 +213,7 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
maxAccountSwitches := h.maxAccountSwitches
switchCount := 0
failedAccountIDs := make(map[int64]struct{})
sameAccountRetryCount := make(map[int64]int)
var lastFailoverErr *service.UpstreamFailoverError
for {
@@ -259,6 +261,7 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
zap.Float64("load_skew", scheduleDecision.LoadSkew),
)
account := selection.Account
sessionHash = ensureOpenAIPoolModeSessionHash(sessionHash, account)
reqLog.Debug("openai.account_selected", zap.Int64("account_id", account.ID), zap.String("account_name", account.Name))
setOpsSelectedAccount(c, account.ID, account.Platform)
@@ -288,6 +291,25 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
var failoverErr *service.UpstreamFailoverError
if errors.As(err, &failoverErr) {
h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, false, nil)
// 池模式:同账号重试
if failoverErr.RetryableOnSameAccount {
retryLimit := account.GetPoolModeRetryCount()
if sameAccountRetryCount[account.ID] < retryLimit {
sameAccountRetryCount[account.ID]++
reqLog.Warn("openai.pool_mode_same_account_retry",
zap.Int64("account_id", account.ID),
zap.Int("upstream_status", failoverErr.StatusCode),
zap.Int("retry_limit", retryLimit),
zap.Int("retry_count", sameAccountRetryCount[account.ID]),
)
select {
case <-c.Request.Context().Done():
return
case <-time.After(sameAccountRetryDelay):
}
continue
}
}
h.gatewayService.RecordOpenAIAccountSwitch()
failedAccountIDs[account.ID] = struct{}{}
lastFailoverErr = failoverErr
@@ -541,6 +563,7 @@ func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
maxAccountSwitches := h.maxAccountSwitches
switchCount := 0
failedAccountIDs := make(map[int64]struct{})
sameAccountRetryCount := make(map[int64]int)
var lastFailoverErr *service.UpstreamFailoverError
for {
@@ -602,6 +625,7 @@ func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
return
}
account := selection.Account
sessionHash = ensureOpenAIPoolModeSessionHash(sessionHash, account)
reqLog.Debug("openai_messages.account_selected", zap.Int64("account_id", account.ID), zap.String("account_name", account.Name))
_ = scheduleDecision
setOpsSelectedAccount(c, account.ID, account.Platform)
@@ -641,6 +665,25 @@ func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
var failoverErr *service.UpstreamFailoverError
if errors.As(err, &failoverErr) {
h.gatewayService.ReportOpenAIAccountScheduleResult(account.ID, false, nil)
// 池模式:同账号重试
if failoverErr.RetryableOnSameAccount {
retryLimit := account.GetPoolModeRetryCount()
if sameAccountRetryCount[account.ID] < retryLimit {
sameAccountRetryCount[account.ID]++
reqLog.Warn("openai_messages.pool_mode_same_account_retry",
zap.Int64("account_id", account.ID),
zap.Int("upstream_status", failoverErr.StatusCode),
zap.Int("retry_limit", retryLimit),
zap.Int("retry_count", sameAccountRetryCount[account.ID]),
)
select {
case <-c.Request.Context().Done():
return
case <-time.After(sameAccountRetryDelay):
}
continue
}
}
h.gatewayService.RecordOpenAIAccountSwitch()
failedAccountIDs[account.ID] = struct{}{}
lastFailoverErr = failoverErr
@@ -1456,6 +1499,14 @@ func setOpenAIClientTransportWS(c *gin.Context) {
service.SetOpenAIClientTransport(c, service.OpenAIClientTransportWS)
}
func ensureOpenAIPoolModeSessionHash(sessionHash string, account *service.Account) string {
if sessionHash != "" || account == nil || !account.IsPoolMode() {
return sessionHash
}
// 为当前请求生成一次性粘性会话键,确保同账号重试不会重新负载均衡到其他账号。
return "openai-pool-retry-" + uuid.NewString()
}
func openAIWSIngressFallbackSessionSeed(userID, apiKeyID int64, groupID *int64) string {
gid := int64(0)
if groupID != nil {