opensource/sub2api
3
0
Fork 0
mirror of https://github.com/Wei-Shaw/sub2api.git synced 2026-09-21 14:19:18 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix(images): pass content-moderation refusals through as 400 instead of retrying

Browse Source 
When gpt-image upstream returns response.completed with no image but a text
refusal (content moderation, e.g. the model replies "this request was judged
unsafe to generate"), the soft-failure path treated it as a probabilistic
upstream failure and returned a retryable UpstreamFailoverError (502). Retrying
or switching accounts is futile for a content-policy block — it just burns
other accounts' quota and still surfaces an opaque 502 to the client.

Distinguish the two no-image cases:
  (A) model text refusal  -> 400 content_policy_violation, no retry, refusal
      reason passed through to the client.
  (B) truly empty response -> unchanged: retryable UpstreamFailoverError (502).

Adds extractOpenAIImagesModelRefusal (extracts the refusal text from
output_text.delta / message output_text, capped at 600 chars) and two unit
tests; the existing empty-response retry test is unaffected.
This commit is contained in:
404QAQ
2026-06-23 16:57:16 +08:00
parent 6936687870
commit 9491de0a39
2 changed files with 128 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/internal/service/openai_images_incomplete_test.go
+47
View File
@@ -121,3 +121,50 @@ func TestImagesOAuthNonStreaming_CompletedNoImageTriggersSameAccountRetry(t *tes
t.Fatal("soft-failure should prefer same-account retry (probabilistic upstream failure)")
}
}
// 内容审核拒绝（模型未出图但输出文字拒绝）应返回 400 content_policy 错误且不重试，
// 而非可重试的 UpstreamFailoverError。
func TestImagesOAuthNonStreaming_ContentRefusalReturns400NoRetry(t *testing.T) {
// 上游：completed 无图，但模型输出了文字拒绝（内容审核场景）。
upstreamSSE := "event: response.created\n" +
"data: {\"type\":\"response.created\",\"response\":{\"id\":\"r\",\"status\":\"in_progress\",\"model\":\"gpt-5.4-mini\",\"output\":[]}}\n\n" +
"event: response.output_text.delta\n" +
"data: {\"type\":\"response.output_text.delta\",\"delta\":\"抱歉，这个请求因涉及违规内容被安全系统判定为不适合生成。\"}\n\n" +
"event: response.completed\n" +
"data: {\"type\":\"response.completed\",\"response\":{\"id\":\"r\",\"status\":\"completed\",\"model\":\"gpt-5.4-mini\",\"output\":[{\"type\":\"message\",\"content\":[{\"type\":\"output_text\",\"text\":\"抱歉，这个请求因涉及违规内容被安全系统判定为不适合生成。\"}]}],\"tool_usage\":{\"image_gen\":{\"output_tokens\":0}}}}\n\n"
rec := httptest.NewRecorder()
c, _ := gin.CreateTestContext(rec)
c.Request = httptest.NewRequest(http.MethodPost, "/v1/images/generations", nil)
resp := &http.Response{StatusCode: http.StatusOK, Header: http.Header{}, Body: io.NopCloser(strings.NewReader(upstreamSSE))}
svc := &OpenAIGatewayService{}
_, _, _, err := svc.handleOpenAIImagesOAuthNonStreamingResponse(resp, c, "b64_json", "gpt-image-2")
if err == nil {
t.Fatal("content refusal should return an error")
}
// 应是不可重试的内容策略错误（400），而非 UpstreamFailoverError。
var failoverErr *UpstreamFailoverError
if errors.As(err, &failoverErr) {
t.Fatalf("content refusal must NOT be a retryable failover error, got %v", failoverErr)
}
var imgErr *OpenAIImagesUpstreamError
if !errors.As(err, &imgErr) {
t.Fatalf("expected *OpenAIImagesUpstreamError, got %T: %v", err, err)
}
if imgErr.StatusCode != http.StatusBadRequest {
t.Fatalf("content refusal should be 400, got %d", imgErr.StatusCode)
}
if !strings.Contains(imgErr.Message, "安全系统") && !strings.Contains(imgErr.Message, "违规") {
t.Fatalf("refusal message should carry model's reason, got %q", imgErr.Message)
}
}
// extractOpenAIImagesModelRefusal：真空响应（无文字）返回空串。
func TestExtractModelRefusal_EmptyWhenNoText(t *testing.T) {
body := "data: {\"type\":\"response.completed\",\"response\":{\"output\":[],\"tool_usage\":{\"image_gen\":{\"output_tokens\":0}}}}\n\n"
if r := extractOpenAIImagesModelRefusal([]byte(body)); r != "" {
t.Fatalf("empty response should yield no refusal, got %q", r)
}
}
backend/internal/service/openai_images_responses.go
+81 -8
View File
@@ -607,6 +607,62 @@ func openAIImagesUpstreamErrorFromSSEPayload(payload []byte) *OpenAIImagesUpstre
}
}
// extractOpenAIImagesModelRefusal 从上游 SSE 响应体提取「模型未出图、改用文字拒绝」
// 的拒绝文本（内容审核场景）。
//
// 上游 response.completed 无图时，模型常以 output_text / message 形式输出拒绝说明
// （如“被安全系统判定为不适合生成”）。这类失败是内容策略拦截，重试/换账号均无效，
// 应把该文本作为内容策略错误透传给客户端。返回空串表示无文字输出（真空响应）。
func extractOpenAIImagesModelRefusal(body []byte) string {
var b strings.Builder
collect := func(s string) {
if s = strings.TrimSpace(s); s != "" {
if b.Len() > 0 {
_ = b.WriteByte(' ')
}
_, _ = b.WriteString(s)
}
}
forEachOpenAISSEDataPayload(string(body), func(payload []byte) {
if !gjson.ValidBytes(payload) {
return
}
switch gjson.GetBytes(payload, "type").String() {
case "response.output_text.delta":
// 流式文本增量。
collect(gjson.GetBytes(payload, "delta").String())
case "response.completed", "response.output_item.done":
// 终态里的 message/output_text。
gjson.GetBytes(payload, "response.output").ForEach(func(_, item gjson.Result) bool {
if item.Get("type").String() == "message" {
item.Get("content").ForEach(func(_, part gjson.Result) bool {
if part.Get("type").String() == "output_text" {
collect(part.Get("text").String())
}
return true
})
}
return true
})
if item := gjson.GetBytes(payload, "item"); item.Get("type").String() == "message" {
item.Get("content").ForEach(func(_, part gjson.Result) bool {
if part.Get("type").String() == "output_text" {
collect(part.Get("text").String())
}
return true
})
}
}
})
refusal := strings.TrimSpace(b.String())
// 截断过长文本，避免把整段模型输出塞进错误响应。
const maxRefusal = 600
if len(refusal) > maxRefusal {
refusal = refusal[:maxRefusal]
}
return refusal
}
// summarizeOpenAIImagesNoOutputBody 从上游 SSE 响应体提取诊断摘要，用于软失败时
// 记录到 ops 日志（上游无图、无标准错误的场景）。提取最终事件类型、response.status、
// incomplete_details.reason，并附 body 截断片段，便于事后定位上游到底返回了什么。
@@ -1037,14 +1093,31 @@ func (s *OpenAIGatewayService) handleOpenAIImagesOAuthNonStreamingResponse(
}
return OpenAIUsage{}, 0, nil, upstreamErr
}
// 软失败兜底：上游既无图、又无任何可识别的 error/failed/incomplete 事件
// （实测：上游偶发把请求路由到 gpt-5.x-mini，返回 response.completed 但 output 为空、
// image_gen 工具未执行）。这是上游的概率性失败——同账号有时成功有时失败。
// 处理：① 记录上游诊断摘要到 ops（last_event/status/model/body 片段）便于排查；
// ② 返回 UpstreamFailoverError 触发重试。因实测为「同账号概率性失败」，优先
// RetryableOnSameAccount 同账号快速重试（默认 3 次，大概率某次正常出图），
// 用尽后由 handler 自然换账号 failover（switchCount 上限保护），既提高成功率
// 又不无谓消耗其他账号配额。
// 软失败兜底：上游无图。先区分两种情形（实测真因，见下）：
//
// (A) 内容审核拒绝：模型未出图，但输出了文字拒绝（response.completed 里带
// output_text / message，内容如“被安全系统判定为不适合生成”）。这是用户
// prompt 触发 OpenAI 内容策略，模型主动拒绝改用文字回应。**换账号/重试均无效**
// （内容层拦截，与账号/承载模型无关），应把拒绝理由作为 400 透传给客户端，
// 避免无谓地重试 + 消耗其它账号配额，且让客户端拿到可读的拒绝原因。
// (B) 真空响应：既无图也无任何文字输出（罕见，如偶发路由到 gpt-5.x-mini、
// image_gen 工具未执行）。这是上游的概率性失败，此时才按可重试处理。
if refusal := extractOpenAIImagesModelRefusal(body); refusal != "" {
refusalErr := &OpenAIImagesUpstreamError{
StatusCode: http.StatusBadRequest,
ErrorType: "image_generation_user_error",
Code: "content_policy_violation",
Message: sanitizeUpstreamErrorMessage(refusal),
}
setOpsUpstreamError(c, http.StatusBadRequest, refusalErr.clientMessage(), summarizeOpenAIImagesNoOutputBody(body))
writeOpenAIImagesUpstreamErrorResponse(c, refusalErr)
return OpenAIUsage{}, 0, nil, refusalErr
}
// (B) 真空响应：记录上游诊断摘要到 ops（last_event/status/model/body 片段）便于
// 排查，并返回 UpstreamFailoverError 触发重试。因实测为「同账号概率性失败」，优先
// RetryableOnSameAccount 同账号快速重试（默认 3 次，大概率某次正常出图），用尽后
// 由 handler 自然换账号 failover（switchCount 上限保护），既提高成功率又不无谓
// 消耗其它账号配额。
setOpsUpstreamError(c, http.StatusBadGateway, "upstream did not return image output", summarizeOpenAIImagesNoOutputBody(body))
return OpenAIUsage{}, 0, nil, &UpstreamFailoverError{
StatusCode: http.StatusBadGateway,