diff --git a/backend/internal/service/antigravity_gateway_service.go b/backend/internal/service/antigravity_gateway_service.go index 81a1c149..71dee705 100644 --- a/backend/internal/service/antigravity_gateway_service.go +++ b/backend/internal/service/antigravity_gateway_service.go @@ -1285,6 +1285,28 @@ func (s *AntigravityGatewayService) Forward(ctx context.Context, c *gin.Context, s.handleUpstreamError(ctx, prefix, account, resp.StatusCode, resp.Header, respBody, originalModel, 0, "", isStickySession) + // 精确匹配服务端配置类 400 错误，触发 failover + 临时封禁 + if resp.StatusCode == http.StatusBadRequest { + msg := strings.ToLower(strings.TrimSpace(extractAntigravityErrorMessage(respBody))) + if isGoogleProjectConfigError(msg) { + upstreamMsg := sanitizeUpstreamErrorMessage(strings.TrimSpace(extractAntigravityErrorMessage(respBody))) + upstreamDetail := s.getUpstreamErrorDetail(respBody) + log.Printf("%s status=400 google_config_error failover=true upstream_message=%q account=%d", prefix, upstreamMsg, account.ID) + appendOpsUpstreamError(c, OpsUpstreamErrorEvent{ + Platform: account.Platform, + AccountID: account.ID, + AccountName: account.Name, + UpstreamStatusCode: resp.StatusCode, + UpstreamRequestID: resp.Header.Get("x-request-id"), + Kind: "failover", + Message: upstreamMsg, + Detail: upstreamDetail, + }) + tempUnscheduleGoogleConfigError(ctx, s.accountRepo, account.ID, prefix) + return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode, ResponseBody: respBody} + } + } + if s.shouldFailoverUpstreamError(resp.StatusCode) { upstreamMsg := strings.TrimSpace(extractAntigravityErrorMessage(respBody)) upstreamMsg = sanitizeUpstreamErrorMessage(upstreamMsg) @@ -1825,6 +1847,23 @@ func (s *AntigravityGatewayService) ForwardGemini(ctx context.Context, c *gin.Co // Always record upstream context for Ops error logs, even when we will failover. setOpsUpstreamError(c, resp.StatusCode, upstreamMsg, upstreamDetail) + // 精确匹配服务端配置类 400 错误，触发 failover + 临时封禁 + if resp.StatusCode == http.StatusBadRequest && isGoogleProjectConfigError(strings.ToLower(upstreamMsg)) { + log.Printf("%s status=400 google_config_error failover=true upstream_message=%q account=%d", prefix, upstreamMsg, account.ID) + appendOpsUpstreamError(c, OpsUpstreamErrorEvent{ + Platform: account.Platform, + AccountID: account.ID, + AccountName: account.Name, + UpstreamStatusCode: resp.StatusCode, + UpstreamRequestID: requestID, + Kind: "failover", + Message: upstreamMsg, + Detail: upstreamDetail, + }) + tempUnscheduleGoogleConfigError(ctx, s.accountRepo, account.ID, prefix) + return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode, ResponseBody: unwrappedForOps} + } + if s.shouldFailoverUpstreamError(resp.StatusCode) { appendOpsUpstreamError(c, OpsUpstreamErrorEvent{ Platform: account.Platform, @@ -1920,6 +1959,29 @@ func (s *AntigravityGatewayService) shouldFailoverUpstreamError(statusCode int) } } +// isGoogleProjectConfigError 判断（已提取的小写）错误消息是否属于 Google 服务端配置类问题。 +// 只精确匹配已知的服务端侧错误，避免对客户端请求错误做无意义重试。 +// 适用于所有走 Google 后端的平台（Antigravity、Gemini）。 +func isGoogleProjectConfigError(lowerMsg string) bool { + // Google 间歇性 Bug：Project ID 有效但被临时识别失败 + return strings.Contains(lowerMsg, "invalid project resource name") +} + +// googleConfigErrorCooldown 服务端配置类 400 错误的临时封禁时长 +const googleConfigErrorCooldown = 60 * time.Minute + +// tempUnscheduleGoogleConfigError 对服务端配置类 400 错误触发临时封禁， +// 避免短时间内反复调度到同一个有问题的账号。 +func tempUnscheduleGoogleConfigError(ctx context.Context, repo AccountRepository, accountID int64, logPrefix string) { + until := time.Now().Add(googleConfigErrorCooldown) + reason := "400: invalid project resource name (auto temp-unschedule 1h)" + if err := repo.SetTempUnschedulable(ctx, accountID, until, reason); err != nil { + log.Printf("%s temp_unschedule_failed account=%d error=%v", logPrefix, accountID, err) + } else { + log.Printf("%s temp_unscheduled account=%d until=%v reason=%q", logPrefix, accountID, until.Format("15:04:05"), reason) + } +} + // sleepAntigravityBackoffWithContext 带 context 取消检查的退避等待 // 返回 true 表示正常完成等待，false 表示 context 已取消 func sleepAntigravityBackoffWithContext(ctx context.Context, attempt int) bool { diff --git a/backend/internal/service/gemini_messages_compat_service.go b/backend/internal/service/gemini_messages_compat_service.go index 792c8f4b..1e59c5fd 100644 --- a/backend/internal/service/gemini_messages_compat_service.go +++ b/backend/internal/service/gemini_messages_compat_service.go @@ -880,6 +880,38 @@ func (s *GeminiMessagesCompatService) Forward(ctx context.Context, c *gin.Contex // ErrorPolicyNone → 原有逻辑 s.handleGeminiUpstreamError(ctx, account, resp.StatusCode, resp.Header, respBody) + // 精确匹配服务端配置类 400 错误，触发 failover + 临时封禁 + if resp.StatusCode == http.StatusBadRequest { + msg400 := strings.ToLower(strings.TrimSpace(extractUpstreamErrorMessage(respBody))) + if isGoogleProjectConfigError(msg400) { + upstreamReqID := resp.Header.Get(requestIDHeader) + if upstreamReqID == "" { + upstreamReqID = resp.Header.Get("x-goog-request-id") + } + upstreamMsg := sanitizeUpstreamErrorMessage(strings.TrimSpace(extractUpstreamErrorMessage(respBody))) + upstreamDetail := "" + if s.cfg != nil && s.cfg.Gateway.LogUpstreamErrorBody { + maxBytes := s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes + if maxBytes <= 0 { + maxBytes = 2048 + } + upstreamDetail = truncateString(string(respBody), maxBytes) + } + log.Printf("[Gemini] status=400 google_config_error failover=true upstream_message=%q account=%d", upstreamMsg, account.ID) + appendOpsUpstreamError(c, OpsUpstreamErrorEvent{ + Platform: account.Platform, + AccountID: account.ID, + AccountName: account.Name, + UpstreamStatusCode: resp.StatusCode, + UpstreamRequestID: upstreamReqID, + Kind: "failover", + Message: upstreamMsg, + Detail: upstreamDetail, + }) + tempUnscheduleGoogleConfigError(ctx, s.accountRepo, account.ID, "[Gemini]") + return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode, ResponseBody: respBody} + } + } if s.shouldFailoverGeminiUpstreamError(resp.StatusCode) { upstreamReqID := resp.Header.Get(requestIDHeader) if upstreamReqID == "" { @@ -1330,6 +1362,35 @@ func (s *GeminiMessagesCompatService) ForwardNative(ctx context.Context, c *gin. // ErrorPolicyNone → 原有逻辑 s.handleGeminiUpstreamError(ctx, account, resp.StatusCode, resp.Header, respBody) + // 精确匹配服务端配置类 400 错误，触发 failover + 临时封禁 + if resp.StatusCode == http.StatusBadRequest { + msg400 := strings.ToLower(strings.TrimSpace(extractUpstreamErrorMessage(respBody))) + if isGoogleProjectConfigError(msg400) { + evBody := unwrapIfNeeded(isOAuth, respBody) + upstreamMsg := sanitizeUpstreamErrorMessage(strings.TrimSpace(extractUpstreamErrorMessage(evBody))) + upstreamDetail := "" + if s.cfg != nil && s.cfg.Gateway.LogUpstreamErrorBody { + maxBytes := s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes + if maxBytes <= 0 { + maxBytes = 2048 + } + upstreamDetail = truncateString(string(evBody), maxBytes) + } + log.Printf("[Gemini] status=400 google_config_error failover=true upstream_message=%q account=%d", upstreamMsg, account.ID) + appendOpsUpstreamError(c, OpsUpstreamErrorEvent{ + Platform: account.Platform, + AccountID: account.ID, + AccountName: account.Name, + UpstreamStatusCode: resp.StatusCode, + UpstreamRequestID: requestID, + Kind: "failover", + Message: upstreamMsg, + Detail: upstreamDetail, + }) + tempUnscheduleGoogleConfigError(ctx, s.accountRepo, account.ID, "[Gemini]") + return nil, &UpstreamFailoverError{StatusCode: resp.StatusCode, ResponseBody: evBody} + } + } if s.shouldFailoverGeminiUpstreamError(resp.StatusCode) { evBody := unwrapIfNeeded(isOAuth, respBody) upstreamMsg := strings.TrimSpace(extractUpstreamErrorMessage(evBody))