diff --git a/sdk/api/handlers/openai/openai_responses_handlers.go b/sdk/api/handlers/openai/openai_responses_handlers.go index 9aa1a497..83eb8bf5 100644 --- a/sdk/api/handlers/openai/openai_responses_handlers.go +++ b/sdk/api/handlers/openai/openai_responses_handlers.go @@ -179,10 +179,9 @@ func (h *OpenAIResponsesAPIHandler) handleStreamingResponse(c *gin.Context, rawJ nonStreamReq := forceResponsesNonStreaming(rawJSON) resp, errMsg := h.ExecuteWithAuthManager(cliCtx, h.HandlerType(), modelName, nonStreamReq, "") if errMsg != nil { - // Factory/Stainless clients often set stream:true; when we fail before emitting any SSE, - // return a JSON error so clients can surface the message (instead of "400 no body"). - c.Header("Content-Type", "application/json") - h.WriteErrorResponse(c, errMsg) + // Factory/Stainless clients request stream:true and may not surface non-2xx JSON bodies well. + // Emit a synthetic SSE response with an assistant-facing error message to avoid "no body". + h.writeSyntheticResponsesErrorSSE(c, flusher, errMsg) cliCancel(errMsg.Error) return } @@ -209,6 +208,50 @@ func (h *OpenAIResponsesAPIHandler) handleStreamingResponse(c *gin.Context, rawJ return } +func (h *OpenAIResponsesAPIHandler) writeSyntheticResponsesErrorSSE(c *gin.Context, flusher http.Flusher, errMsg *interfaces.ErrorMessage) { + // Always emit SSE with HTTP 200 so strict streaming clients can display the message. + c.Status(http.StatusOK) + c.Header("Content-Type", "text/event-stream") + c.Header("Cache-Control", "no-cache") + c.Header("Connection", "keep-alive") + c.Header("Access-Control-Allow-Origin", "*") + + msg := "Request failed." + statusCode := 0 + if errMsg != nil { + statusCode = errMsg.StatusCode + if errMsg.Error != nil && errMsg.Error.Error() != "" { + raw := errMsg.Error.Error() + // Try to extract a user-facing message from common upstream error shapes. + if gjson.Valid(raw) { + if v := gjson.Get(raw, "error.message"); v.Exists() && v.String() != "" { + msg = v.String() + } else if v := gjson.Get(raw, "error.error.message"); v.Exists() && v.String() != "" { + msg = v.String() + } else if v := gjson.Get(raw, "message"); v.Exists() && v.String() != "" { + msg = v.String() + } else { + msg = raw + } + } else { + msg = raw + } + } + } + if statusCode > 0 { + msg = fmt.Sprintf("[HTTP %d] %s", statusCode, strings.TrimSpace(msg)) + } + + // Keep the message short; some clients stuff huge upstream JSON error blobs. + const maxChars = 1800 + if len(msg) > maxChars { + msg = msg[:maxChars] + "\n...[truncated]..." + } + + resp := fmt.Sprintf(`{"id":"resp_error_%d","output_text":%q}`, time.Now().UnixNano(), msg) + h.writeSyntheticResponsesSSE(c, flusher, []byte(resp)) +} + func (h *OpenAIResponsesAPIHandler) forwardResponsesStream(c *gin.Context, flusher http.Flusher, cancel func(error), data <-chan []byte, errs <-chan *interfaces.ErrorMessage) { for { select {