Merge pull request #4180 from jianjianai/fix/openai-request-view-single-scan

fix(openai): avoid repeated request body scans

backend/internal/service/openai_gateway_request_body.go

@@ -365,15 +365,57 @@ func newOpenAIRequestView(body []byte) openAIRequestView {
 	if len(body) == 0 {
 		return openAIRequestView{}
 	}
-	return openAIRequestView{
-		body:               body,
-		Model:              strings.TrimSpace(gjson.GetBytes(body, "model").String()),
-		Stream:             gjson.GetBytes(body, "stream").Bool(),
-		PromptCacheKey:     strings.TrimSpace(gjson.GetBytes(body, "prompt_cache_key").String()),
-		PreviousResponseID: strings.TrimSpace(gjson.GetBytes(body, "previous_response_id").String()),
-		ServiceTier:        strings.TrimSpace(gjson.GetBytes(body, "service_tier").String()),
-		ReasoningEffort:    strings.TrimSpace(gjson.GetBytes(body, "reasoning.effort").String()),
-	}
+
+	const (
+		modelField uint8 = 1 << iota
+		streamField
+		promptCacheKeyField
+		previousResponseIDField
+		serviceTierField
+		reasoningField
+		allRequestViewFields = modelField | streamField | promptCacheKeyField |
+			previousResponseIDField | serviceTierField | reasoningField
+	)
+
+	view := openAIRequestView{body: body}
+	var seen uint8
+	// parseRawJSONView reads body without copying; view keeps body alive for extracted strings.
+	parseRawJSONView(body).ForEach(func(key, value gjson.Result) bool {
+		switch key.Str {
+		case "model":
+			if seen&modelField == 0 {
+				view.Model = strings.TrimSpace(value.String())
+				seen |= modelField
+			}
+		case "stream":
+			if seen&streamField == 0 {
+				view.Stream = value.Bool()
+				seen |= streamField
+			}
+		case "prompt_cache_key":
+			if seen&promptCacheKeyField == 0 {
+				view.PromptCacheKey = strings.TrimSpace(value.String())
+				seen |= promptCacheKeyField
+			}
+		case "previous_response_id":
+			if seen&previousResponseIDField == 0 {
+				view.PreviousResponseID = strings.TrimSpace(value.String())
+				seen |= previousResponseIDField
+			}
+		case "service_tier":
+			if seen&serviceTierField == 0 {
+				view.ServiceTier = strings.TrimSpace(value.String())
+				seen |= serviceTierField
+			}
+		case "reasoning":
+			if seen&reasoningField == 0 {
+				view.ReasoningEffort = strings.TrimSpace(value.Get("effort").String())
+				seen |= reasoningField
+			}
+		}
+		return seen != allRequestViewFields
+	})
+	return view
 }
 
 // Decode 保留阶段一既有 full-map 行为；后续阶段会把调用点下沉到复杂分支。

backend/internal/service/openai_gateway_service_hotpath_test.go

@@ -27,6 +27,33 @@ func TestOpenAIRequestView_ExtractsRawScalars(t *testing.T) {
 	require.Equal(t, "medium", view.ReasoningEffort)
 }
 
+func TestOpenAIRequestView_ExtractsFieldsAfterLargeInput(t *testing.T) {
+	body := []byte(`{"model":"gpt-5","input":[{"content":"` + strings.Repeat("payload", 1024) + `"}],"stream":true,"prompt_cache_key":"session-1","previous_response_id":"resp-1","service_tier":"flex","reasoning":{"effort":"high"}}`)
+
+	view := newOpenAIRequestView(body)
+
+	require.Equal(t, "gpt-5", view.Model)
+	require.True(t, view.Stream)
+	require.Equal(t, "session-1", view.PromptCacheKey)
+	require.Equal(t, "resp-1", view.PreviousResponseID)
+	require.Equal(t, "flex", view.ServiceTier)
+	require.Equal(t, "high", view.ReasoningEffort)
+}
+
+func TestOpenAIRequestView_KeepsFirstDuplicateField(t *testing.T) {
+	view := newOpenAIRequestView([]byte(`{"model":"gpt-5","model":"gpt-5.1","reasoning":{"effort":"low"},"reasoning":{"effort":"high"}}`))
+
+	require.Equal(t, "gpt-5", view.Model)
+	require.Equal(t, "low", view.ReasoningEffort)
+}
+
+func TestOpenAIRequestView_KeepsLenientPrefixExtraction(t *testing.T) {
+	view := newOpenAIRequestView([]byte(`{"model":"gpt-5","stream":true,"input":[`))
+
+	require.Equal(t, "gpt-5", view.Model)
+	require.True(t, view.Stream)
+}
+
 func TestOpenAIRequestView_DecodeKeepsFullMapBehavior(t *testing.T) {
 	view := newOpenAIRequestView([]byte(`{"model":"gpt-5","stream":true,"input":[{"type":"message","content":"hi"}]}`))