oadmin/sub2api
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

refactor: extract failover error handling into FailoverState

Browse Source 
- Extract duplicated failover logic from gateway_handler.go (3 places)
  and gemini_v1beta_handler.go into shared failover_loop.go
- Introduce FailoverState with HandleFailoverError and HandleSelectionExhausted
- Move helper functions (needForceCacheBilling, sleepWithContext) into failover_loop.go
- Add comprehensive unit tests (32+ test cases)
- Delete redundant gateway_handler_single_account_retry_test.go
This commit is contained in:
erio
2026-02-24 18:08:04 +08:00
parent aaac1aaca9
commit 09166a52f8
5 changed files with 975 additions and 303 deletions
Download Patch File Download Diff File Expand all files Collapse all files

75
backend/internal/handler/gemini_v1beta_handler.go
View File

@@ -344,11 +344,7 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
hasBoundSession := sessionKey != "" && sessionBoundAccountID > 0
cleanedForUnknownBinding := false
maxAccountSwitches := h.maxAccountSwitchesGemini
switchCount := 0
failedAccountIDs := make(map[int64]struct{})
var lastFailoverErr *service.UpstreamFailoverError
var forceCacheBilling bool // 粘性会话切换时的缓存计费标记
fs := NewFailoverState(h.maxAccountSwitchesGemini, hasBoundSession)
// 单账号分组提前设置 SingleAccountRetry 标记，让 Service 层首次 503 就不设模型限流标记。
// 避免单账号分组收到 503 (MODEL_CAPACITY_EXHAUSTED) 时设 29s 限流，导致后续请求连续快速失败。
@@ -358,30 +354,24 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
}
for {
selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionKey, modelName, failedAccountIDs, "") // Gemini 不使用会话限制
selection, err := h.gatewayService.SelectAccountWithLoadAwareness(c.Request.Context(), apiKey.GroupID, sessionKey, modelName, fs.FailedAccountIDs, "") // Gemini 不使用会话限制
if err != nil {
if len(failedAccountIDs) == 0 {
if len(fs.FailedAccountIDs) == 0 {
googleError(c, http.StatusServiceUnavailable, "No available Gemini accounts: "+err.Error())
return
}
// Antigravity 单账号退避重试：分组内没有其他可用账号时，
// 对 503 错误不直接返回，而是清除排除列表、等待退避后重试同一个账号。
// 谷歌上游 503 (MODEL_CAPACITY_EXHAUSTED) 通常是暂时性的，等几秒就能恢复。
if lastFailoverErr != nil && lastFailoverErr.StatusCode == http.StatusServiceUnavailable && switchCount <= maxAccountSwitches {
if sleepAntigravitySingleAccountBackoff(c.Request.Context(), switchCount) {
reqLog.Warn("gemini.single_account_retrying",
zap.Int("retry_count", switchCount),
zap.Int("max_retries", maxAccountSwitches),
)
failedAccountIDs = make(map[int64]struct{})
// 设置 context 标记，让 Service 层预检查等待限流过期而非直接切换
ctx := context.WithValue(c.Request.Context(), ctxkey.SingleAccountRetry, true)
c.Request = c.Request.WithContext(ctx)
continue
}
action := fs.HandleSelectionExhausted(c.Request.Context())
switch action {
case FailoverContinue:
ctx := context.WithValue(c.Request.Context(), ctxkey.SingleAccountRetry, true)
c.Request = c.Request.WithContext(ctx)
continue
case FailoverCanceled:
return
default: // FailoverExhausted
h.handleGeminiFailoverExhausted(c, fs.LastFailoverErr)
return
}
h.handleGeminiFailoverExhausted(c, lastFailoverErr)
return
}
account := selection.Account
setOpsSelectedAccount(c, account.ID, account.Platform)
@@ -465,8 +455,8 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
// 5) forward (根据平台分流)
var result *service.ForwardResult
requestCtx := c.Request.Context()
if switchCount > 0 {
requestCtx = context.WithValue(requestCtx, ctxkey.AccountSwitchCount, switchCount)
if fs.SwitchCount > 0 {
requestCtx = context.WithValue(requestCtx, ctxkey.AccountSwitchCount, fs.SwitchCount)
}
if account.Platform == service.PlatformAntigravity && account.Type != service.AccountTypeAPIKey {
result, err = h.antigravityGatewayService.ForwardGemini(requestCtx, c, account, modelName, action, stream, body, hasBoundSession)
@@ -479,29 +469,16 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
if err != nil {
var failoverErr *service.UpstreamFailoverError
if errors.As(err, &failoverErr) {
failedAccountIDs[account.ID] = struct{}{}
if needForceCacheBilling(hasBoundSession, failoverErr) {
forceCacheBilling = true
}
if switchCount >= maxAccountSwitches {
lastFailoverErr = failoverErr
h.handleGeminiFailoverExhausted(c, lastFailoverErr)
failoverAction := fs.HandleFailoverError(c.Request.Context(), h.gatewayService, account.ID, account.Platform, failoverErr)
switch failoverAction {
case FailoverContinue:
continue
case FailoverExhausted:
h.handleGeminiFailoverExhausted(c, fs.LastFailoverErr)
return
case FailoverCanceled:
return
}
lastFailoverErr = failoverErr
switchCount++
reqLog.Warn("gemini.upstream_failover_switching",
zap.Int64("account_id", account.ID),
zap.Int("upstream_status", failoverErr.StatusCode),
zap.Int("switch_count", switchCount),
zap.Int("max_switches", maxAccountSwitches),
)
if account.Platform == service.PlatformAntigravity {
if !sleepFailoverDelay(c.Request.Context(), switchCount) {
return
}
}
continue
}
// ForwardNative already wrote the response
reqLog.Error("gemini.forward_failed", zap.Int64("account_id", account.ID), zap.Error(err))
@@ -539,7 +516,7 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
IPAddress: clientIP,
LongContextThreshold: 200000, // Gemini 200K 阈值
LongContextMultiplier: 2.0, // 超出部分双倍计费
ForceCacheBilling: forceCacheBilling,
ForceCacheBilling: fs.ForceCacheBilling,
APIKeyService: h.apiKeyService,
}); err != nil {
logger.L().With(
@@ -554,7 +531,7 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
})
reqLog.Debug("gemini.request_completed",
zap.Int64("account_id", account.ID),
zap.Int("switch_count", switchCount),
zap.Int("switch_count", fs.SwitchCount),
)
return
}