github/sub2api
4
0
Fork 0
mirror of https://github.com/Wei-Shaw/sub2api.git synced 2026-10-28 08:08:32 +00:00
Code Issues Packages Projects Releases Wiki Activity

merge: resolve conflicts with upstream/main

Browse Source 
Keep model-scoped reasoning effort mappings together with the
deny-or-downgrade ceiling control that landed on main.
This commit is contained in:
刘佳林
2026-09-02 10:01:42 +08:00
parent 7c01ec9be8 6566039bc8
commit 77729e2723
179 changed files with 6415 additions and 1152 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/internal/handler/admin/channel_handler.go
+82 -65
View File
@@ -57,20 +57,21 @@ type updateChannelRequest struct {
}
type channelModelPricingRequest struct {
Platform string `json:"platform" binding:"omitempty,max=50"`
Models []string `json:"models" binding:"required,min=1,max=100"`
BillingMode string `json:"billing_mode" binding:"omitempty,oneof=token per_request image"`
InputPrice *float64 `json:"input_price" binding:"omitempty,min=0"`
OutputPrice *float64 `json:"output_price" binding:"omitempty,min=0"`
CacheWritePrice *float64 `json:"cache_write_price" binding:"omitempty,min=0"`
CacheReadPrice *float64 `json:"cache_read_price" binding:"omitempty,min=0"`
FastMultiplier *float64 `json:"fast_multiplier" binding:"omitempty,gt=0"`
FlexMultiplier *float64 `json:"flex_multiplier" binding:"omitempty,gt=0"`
ImageInputPrice *float64 `json:"image_input_price" binding:"omitempty,min=0"`
ImageOutputPrice *float64 `json:"image_output_price" binding:"omitempty,min=0"`
PerRequestPrice *float64 `json:"per_request_price" binding:"omitempty,min=0"`
Intervals []pricingIntervalRequest `json:"intervals"`
TimePricing *channelTimePricingRequest `json:"time_pricing"`
Platform string `json:"platform" binding:"omitempty,max=50"`
Models []string `json:"models" binding:"required,min=1,max=100"`
BillingMode string `json:"billing_mode" binding:"omitempty,oneof=token per_request image"`
InputPrice *float64 `json:"input_price" binding:"omitempty,min=0"`
OutputPrice *float64 `json:"output_price" binding:"omitempty,min=0"`
CacheWritePrice *float64 `json:"cache_write_price" binding:"omitempty,min=0"`
CacheWrite1hPrice *float64 `json:"cache_write_1h_price" binding:"omitempty,min=0"`
CacheReadPrice *float64 `json:"cache_read_price" binding:"omitempty,min=0"`
FastMultiplier *float64 `json:"fast_multiplier" binding:"omitempty,gt=0"`
FlexMultiplier *float64 `json:"flex_multiplier" binding:"omitempty,gt=0"`
ImageInputPrice *float64 `json:"image_input_price" binding:"omitempty,min=0"`
ImageOutputPrice *float64 `json:"image_output_price" binding:"omitempty,min=0"`
PerRequestPrice *float64 `json:"per_request_price" binding:"omitempty,min=0"`
Intervals []pricingIntervalRequest `json:"intervals"`
TimePricing *channelTimePricingRequest `json:"time_pricing"`
}
type channelTimePricingRequest struct {
@@ -92,6 +93,7 @@ type pricingIntervalRequest struct {
InputPrice *float64 `json:"input_price"`
OutputPrice *float64 `json:"output_price"`
CacheWritePrice *float64 `json:"cache_write_price"`
CacheWrite1hPrice *float64 `json:"cache_write_1h_price"`
CacheReadPrice *float64 `json:"cache_read_price"`
InputMultiplier *float64 `json:"input_multiplier" binding:"omitempty,gt=0"`
OutputMultiplier *float64 `json:"output_multiplier" binding:"omitempty,gt=0"`
@@ -127,21 +129,22 @@ type channelResponse struct {
}
type channelModelPricingResponse struct {
ID int64 `json:"id"`
Platform string `json:"platform"`
Models []string `json:"models"`
BillingMode string `json:"billing_mode"`
InputPrice *float64 `json:"input_price"`
OutputPrice *float64 `json:"output_price"`
CacheWritePrice *float64 `json:"cache_write_price"`
CacheReadPrice *float64 `json:"cache_read_price"`
FastMultiplier *float64 `json:"fast_multiplier"`
FlexMultiplier *float64 `json:"flex_multiplier"`
ImageInputPrice *float64 `json:"image_input_price"`
ImageOutputPrice *float64 `json:"image_output_price"`
PerRequestPrice *float64 `json:"per_request_price"`
Intervals []pricingIntervalResponse `json:"intervals"`
TimePricing *channelTimePricingResponse `json:"time_pricing"`
ID int64 `json:"id"`
Platform string `json:"platform"`
Models []string `json:"models"`
BillingMode string `json:"billing_mode"`
InputPrice *float64 `json:"input_price"`
OutputPrice *float64 `json:"output_price"`
CacheWritePrice *float64 `json:"cache_write_price"`
CacheWrite1hPrice *float64 `json:"cache_write_1h_price"`
CacheReadPrice *float64 `json:"cache_read_price"`
FastMultiplier *float64 `json:"fast_multiplier"`
FlexMultiplier *float64 `json:"flex_multiplier"`
ImageInputPrice *float64 `json:"image_input_price"`
ImageOutputPrice *float64 `json:"image_output_price"`
PerRequestPrice *float64 `json:"per_request_price"`
Intervals []pricingIntervalResponse `json:"intervals"`
TimePricing *channelTimePricingResponse `json:"time_pricing"`
}
type channelTimePricingResponse struct {
@@ -164,6 +167,7 @@ type pricingIntervalResponse struct {
InputPrice *float64 `json:"input_price"`
OutputPrice *float64 `json:"output_price"`
CacheWritePrice *float64 `json:"cache_write_price"`
CacheWrite1hPrice *float64 `json:"cache_write_1h_price"`
CacheReadPrice *float64 `json:"cache_read_price"`
InputMultiplier *float64 `json:"input_multiplier"`
OutputMultiplier *float64 `json:"output_multiplier"`
@@ -254,21 +258,22 @@ func pricingToResponse(p *service.ChannelModelPricing) channelModelPricingRespon
intervals = append(intervals, intervalToResponse(iv))
}
return channelModelPricingResponse{
ID: p.ID,
Platform: platform,
Models: models,
BillingMode: billingMode,
InputPrice: p.InputPrice,
OutputPrice: p.OutputPrice,
CacheWritePrice: p.CacheWritePrice,
CacheReadPrice: p.CacheReadPrice,
FastMultiplier: p.FastMultiplier,
FlexMultiplier: p.FlexMultiplier,
ImageInputPrice: p.ImageInputPrice,
ImageOutputPrice: p.ImageOutputPrice,
PerRequestPrice: p.PerRequestPrice,
Intervals: intervals,
TimePricing: timePricingToResponse(p.TimePricing),
ID: p.ID,
Platform: platform,
Models: models,
BillingMode: billingMode,
InputPrice: p.InputPrice,
OutputPrice: p.OutputPrice,
CacheWritePrice: p.CacheWritePrice,
CacheWrite1hPrice: p.CacheWrite1hPrice,
CacheReadPrice: p.CacheReadPrice,
FastMultiplier: p.FastMultiplier,
FlexMultiplier: p.FlexMultiplier,
ImageInputPrice: p.ImageInputPrice,
ImageOutputPrice: p.ImageOutputPrice,
PerRequestPrice: p.PerRequestPrice,
Intervals: intervals,
TimePricing: timePricingToResponse(p.TimePricing),
}
}
@@ -300,6 +305,7 @@ func intervalToResponse(iv service.PricingInterval) pricingIntervalResponse {
InputPrice: iv.InputPrice,
OutputPrice: iv.OutputPrice,
CacheWritePrice: iv.CacheWritePrice,
CacheWrite1hPrice: iv.CacheWrite1hPrice,
CacheReadPrice: iv.CacheReadPrice,
InputMultiplier: iv.InputMultiplier,
OutputMultiplier: iv.OutputMultiplier,
@@ -334,6 +340,7 @@ func pricingRequestToService(reqs []channelModelPricingRequest, allowChannelMult
InputPrice: iv.InputPrice,
OutputPrice: iv.OutputPrice,
CacheWritePrice: iv.CacheWritePrice,
CacheWrite1hPrice: iv.CacheWrite1hPrice,
CacheReadPrice: iv.CacheReadPrice,
InputMultiplier: inputMultiplier,
OutputMultiplier: outputMultiplier,
@@ -349,20 +356,21 @@ func pricingRequestToService(reqs []channelModelPricingRequest, allowChannelMult
flexMultiplier = r.FlexMultiplier
}
result = append(result, service.ChannelModelPricing{
Platform: platform,
Models: r.Models,
BillingMode: billingMode,
InputPrice: r.InputPrice,
OutputPrice: r.OutputPrice,
CacheWritePrice: r.CacheWritePrice,
CacheReadPrice: r.CacheReadPrice,
FastMultiplier: fastMultiplier,
FlexMultiplier: flexMultiplier,
ImageInputPrice: r.ImageInputPrice,
ImageOutputPrice: r.ImageOutputPrice,
PerRequestPrice: r.PerRequestPrice,
Intervals: intervals,
TimePricing: timePricingRequestToService(r.TimePricing),
Platform: platform,
Models: r.Models,
BillingMode: billingMode,
InputPrice: r.InputPrice,
OutputPrice: r.OutputPrice,
CacheWritePrice: r.CacheWritePrice,
CacheWrite1hPrice: r.CacheWrite1hPrice,
CacheReadPrice: r.CacheReadPrice,
FastMultiplier: fastMultiplier,
FlexMultiplier: flexMultiplier,
ImageInputPrice: r.ImageInputPrice,
ImageOutputPrice: r.ImageOutputPrice,
PerRequestPrice: r.PerRequestPrice,
Intervals: intervals,
TimePricing: timePricingRequestToService(r.TimePricing),
})
}
return result
@@ -597,15 +605,24 @@ func (h *ChannelHandler) GetModelDefaultPricing(c *gin.Context) {
response.Success(c, gin.H{"found": false})
return
}
cacheWritePrice := pricing.CacheCreationPricePerToken
var cacheWrite1hPrice *float64
if pricing.SupportsCacheBreakdown {
if pricing.CacheCreation5mPrice > 0 {
cacheWritePrice = pricing.CacheCreation5mPrice
}
cacheWrite1hPrice = &pricing.CacheCreation1hPrice
}
response.Success(c, gin.H{
"found": true,
"input_price": pricing.InputPricePerToken,
"output_price": pricing.OutputPricePerToken,
"cache_write_price": pricing.CacheCreationPricePerToken,
"cache_read_price": pricing.CacheReadPricePerToken,
"image_input_price": pricing.ImageInputPricePerToken,
"image_output_price": pricing.ImageOutputPricePerToken,
"found": true,
"input_price": pricing.InputPricePerToken,
"output_price": pricing.OutputPricePerToken,
"cache_write_price": cacheWritePrice,
"cache_write_1h_price": cacheWrite1hPrice,
"cache_read_price": pricing.CacheReadPricePerToken,
"image_input_price": pricing.ImageInputPricePerToken,
"image_output_price": pricing.ImageOutputPricePerToken,
})
}
backend/internal/handler/admin/channel_handler_test.go
+69 -18
View File
@@ -320,15 +320,16 @@ func TestPricingRequestToService_Defaults(t *testing.T) {
func TestPricingRequestToService_WithAllFields(t *testing.T) {
reqs := []channelModelPricingRequest{
{
Platform: "openai",
Models: []string{"gpt-4", "gpt-4o"},
BillingMode: "per_request",
InputPrice: float64Ptr(0.01),
OutputPrice: float64Ptr(0.03),
CacheWritePrice: float64Ptr(0.005),
CacheReadPrice: float64Ptr(0.002),
ImageOutputPrice: float64Ptr(0.04),
PerRequestPrice: float64Ptr(0.5),
Platform: "openai",
Models: []string{"gpt-4", "gpt-4o"},
BillingMode: "per_request",
InputPrice: float64Ptr(0.01),
OutputPrice: float64Ptr(0.03),
CacheWritePrice: float64Ptr(0.005),
CacheWrite1hPrice: float64Ptr(0.008),
CacheReadPrice: float64Ptr(0.002),
ImageOutputPrice: float64Ptr(0.04),
PerRequestPrice: float64Ptr(0.5),
},
}
@@ -341,6 +342,7 @@ func TestPricingRequestToService_WithAllFields(t *testing.T) {
require.Equal(t, float64Ptr(0.01), r.InputPrice)
require.Equal(t, float64Ptr(0.03), r.OutputPrice)
require.Equal(t, float64Ptr(0.005), r.CacheWritePrice)
require.Equal(t, float64Ptr(0.008), r.CacheWrite1hPrice)
require.Equal(t, float64Ptr(0.002), r.CacheReadPrice)
require.Equal(t, float64Ptr(0.04), r.ImageOutputPrice)
require.Equal(t, float64Ptr(0.5), r.PerRequestPrice)
@@ -353,15 +355,16 @@ func TestPricingRequestToService_WithIntervals(t *testing.T) {
BillingMode: "per_request",
Intervals: []pricingIntervalRequest{
{
MinTokens: 0,
MaxTokens: intPtr(2000),
TierLabel: "small",
InputPrice: float64Ptr(0.01),
OutputPrice: float64Ptr(0.02),
CacheWritePrice: float64Ptr(0.003),
CacheReadPrice: float64Ptr(0.001),
PerRequestPrice: float64Ptr(0.1),
SortOrder: 1,
MinTokens: 0,
MaxTokens: intPtr(2000),
TierLabel: "small",
InputPrice: float64Ptr(0.01),
OutputPrice: float64Ptr(0.02),
CacheWritePrice: float64Ptr(0.003),
CacheWrite1hPrice: float64Ptr(0.006),
CacheReadPrice: float64Ptr(0.001),
PerRequestPrice: float64Ptr(0.1),
SortOrder: 1,
},
{
MinTokens: 2000,
@@ -384,6 +387,7 @@ func TestPricingRequestToService_WithIntervals(t *testing.T) {
require.Equal(t, float64Ptr(0.01), iv0.InputPrice)
require.Equal(t, float64Ptr(0.02), iv0.OutputPrice)
require.Equal(t, float64Ptr(0.003), iv0.CacheWritePrice)
require.Equal(t, float64Ptr(0.006), iv0.CacheWrite1hPrice)
require.Equal(t, float64Ptr(0.001), iv0.CacheReadPrice)
require.Equal(t, float64Ptr(0.1), iv0.PerRequestPrice)
require.Equal(t, 1, iv0.SortOrder)
@@ -558,3 +562,50 @@ func TestSyncPricingModels_ValidPlatform_EmptyService(t *testing.T) {
require.NotNil(t, body.Data.Models, "models must not be null for platform=%s", platform)
}
}
func setupModelDefaultPricingRouter() *gin.Engine {
gin.SetMode(gin.TestMode)
router := gin.New()
h := &ChannelHandler{billingService: service.NewBillingService(nil, nil)}
router.GET("/channels/model-pricing", h.GetModelDefaultPricing)
return router
}
func TestGetModelDefaultPricing_ReturnsFable51CacheTTLs(t *testing.T) {
router := setupModelDefaultPricingRouter()
req := httptest.NewRequest(http.MethodGet, "/channels/model-pricing?model=claude-fable-5-1", nil)
w := httptest.NewRecorder()
router.ServeHTTP(w, req)
require.Equal(t, http.StatusOK, w.Code)
var body struct {
Data struct {
Found bool `json:"found"`
CacheWritePrice float64 `json:"cache_write_price"`
CacheWrite1hPrice *float64 `json:"cache_write_1h_price"`
} `json:"data"`
}
require.NoError(t, json.Unmarshal(w.Body.Bytes(), &body))
require.True(t, body.Data.Found)
require.InDelta(t, 12.5e-6, body.Data.CacheWritePrice, 1e-12)
require.NotNil(t, body.Data.CacheWrite1hPrice)
require.InDelta(t, 20e-6, *body.Data.CacheWrite1hPrice, 1e-12)
}
func TestGetModelDefaultPricing_OmitsUnsupportedCache1hPrice(t *testing.T) {
router := setupModelDefaultPricingRouter()
req := httptest.NewRequest(http.MethodGet, "/channels/model-pricing?model=claude-sonnet-4", nil)
w := httptest.NewRecorder()
router.ServeHTTP(w, req)
require.Equal(t, http.StatusOK, w.Code)
var body struct {
Data struct {
Found bool `json:"found"`
CacheWrite1hPrice *float64 `json:"cache_write_1h_price"`
} `json:"data"`
}
require.NoError(t, json.Unmarshal(w.Body.Bytes(), &body))
require.True(t, body.Data.Found)
require.Nil(t, body.Data.CacheWrite1hPrice)
}
backend/internal/handler/admin/group_handler.go
+14
View File
@@ -147,6 +147,8 @@ type CreateGroupRequest struct {
// OpenAI Messages 调度配置（仅 openai 平台使用）
AllowMessagesDispatch bool `json:"allow_messages_dispatch"`
AllowLive bool `json:"allow_live"`
ForceOpenAIFast bool `json:"force_openai_fast"`
FreeOpenAIFast bool `json:"free_openai_fast"`
RequireOAuthOnly bool `json:"require_oauth_only"`
RequirePrivacySet bool `json:"require_privacy_set"`
DefaultMappedModel string `json:"default_mapped_model"`
@@ -156,6 +158,8 @@ type CreateGroupRequest struct {
RPMLimit int `json:"rpm_limit"`
// OpenAI/Codex 请求推理强度上限，空字符串表示不限制。
MaxReasoningEffort string `json:"max_reasoning_effort"`
// 超过上限时的访问控制：downgrade（默认）或 deny。
MaxReasoningEffortOverLimit string `json:"max_reasoning_effort_over_limit"`
// OpenAI/Codex 推理强度映射，可按模型精确名、前缀或后缀限定。
ReasoningEffortMappings []service.ReasoningEffortMapping `json:"reasoning_effort_mappings"`
// 从指定分组复制账号（创建后自动绑定）
@@ -216,6 +220,8 @@ type UpdateGroupRequest struct {
// OpenAI Messages 调度配置（仅 openai 平台使用）
AllowMessagesDispatch *bool `json:"allow_messages_dispatch"`
AllowLive *bool `json:"allow_live"`
ForceOpenAIFast *bool `json:"force_openai_fast"`
FreeOpenAIFast *bool `json:"free_openai_fast"`
RequireOAuthOnly *bool `json:"require_oauth_only"`
RequirePrivacySet *bool `json:"require_privacy_set"`
DefaultMappedModel *string `json:"default_mapped_model"`
@@ -225,6 +231,8 @@ type UpdateGroupRequest struct {
RPMLimit *int `json:"rpm_limit"`
// OpenAI/Codex 请求推理强度上限；空字符串清除，nil 不修改。
MaxReasoningEffort *string `json:"max_reasoning_effort"`
// 超过上限时的访问控制；空字符串视为 downgrade，nil 不修改。
MaxReasoningEffortOverLimit *string `json:"max_reasoning_effort_over_limit"`
// nil 不修改，空数组清空，非空数组替换。
ReasoningEffortMappings *[]service.ReasoningEffortMapping `json:"reasoning_effort_mappings"`
// 从指定分组复制账号（同步操作：先清空当前分组的账号绑定，再绑定源分组的账号）
@@ -550,6 +558,8 @@ func (h *GroupHandler) Create(c *gin.Context) {
SupportedModelScopes: req.SupportedModelScopes,
AllowMessagesDispatch: req.AllowMessagesDispatch,
AllowLive: req.AllowLive,
ForceOpenAIFast: req.ForceOpenAIFast,
FreeOpenAIFast: req.FreeOpenAIFast,
RequireOAuthOnly: req.RequireOAuthOnly,
RequirePrivacySet: req.RequirePrivacySet,
DefaultMappedModel: req.DefaultMappedModel,
@@ -557,6 +567,7 @@ func (h *GroupHandler) Create(c *gin.Context) {
ModelsListConfig: req.ModelsListConfig,
RPMLimit: req.RPMLimit,
MaxReasoningEffort: req.MaxReasoningEffort,
MaxReasoningEffortOverLimit: req.MaxReasoningEffortOverLimit,
ReasoningEffortMappings: req.ReasoningEffortMappings,
CopyAccountsFromGroupIDs: req.CopyAccountsFromGroupIDs,
})
@@ -679,6 +690,8 @@ func (h *GroupHandler) Update(c *gin.Context) {
SupportedModelScopes: req.SupportedModelScopes,
AllowMessagesDispatch: req.AllowMessagesDispatch,
AllowLive: req.AllowLive,
ForceOpenAIFast: req.ForceOpenAIFast,
FreeOpenAIFast: req.FreeOpenAIFast,
RequireOAuthOnly: req.RequireOAuthOnly,
RequirePrivacySet: req.RequirePrivacySet,
DefaultMappedModel: req.DefaultMappedModel,
@@ -686,6 +699,7 @@ func (h *GroupHandler) Update(c *gin.Context) {
ModelsListConfig: req.ModelsListConfig,
RPMLimit: req.RPMLimit,
MaxReasoningEffort: req.MaxReasoningEffort,
MaxReasoningEffortOverLimit: req.MaxReasoningEffortOverLimit,
ReasoningEffortMappings: req.ReasoningEffortMappings,
CopyAccountsFromGroupIDs: req.CopyAccountsFromGroupIDs,
})
backend/internal/handler/admin/group_handler_openai_fast_test.go
+27
View File
@@ -0,0 +1,27 @@
package admin
import (
"encoding/json"
"testing"
"github.com/stretchr/testify/require"
)
func TestGroupRequestsDecodeForceOpenAIFast(t *testing.T) {
var createReq CreateGroupRequest
require.NoError(t, json.Unmarshal([]byte(`{"name":"fast","force_openai_fast":true,"free_openai_fast":true}`), &createReq))
require.True(t, createReq.ForceOpenAIFast)
require.True(t, createReq.FreeOpenAIFast)
var updateReq UpdateGroupRequest
require.NoError(t, json.Unmarshal([]byte(`{"force_openai_fast":false,"free_openai_fast":false}`), &updateReq))
require.NotNil(t, updateReq.ForceOpenAIFast)
require.False(t, *updateReq.ForceOpenAIFast)
require.NotNil(t, updateReq.FreeOpenAIFast)
require.False(t, *updateReq.FreeOpenAIFast)
var omitted UpdateGroupRequest
require.NoError(t, json.Unmarshal([]byte(`{}`), &omitted))
require.Nil(t, omitted.ForceOpenAIFast)
require.Nil(t, omitted.FreeOpenAIFast)
}
backend/internal/handler/available_channel_handler.go
+38 -34
View File
@@ -66,27 +66,29 @@ type userAvailableGroup struct {
// userSupportedModelPricing 用户可见的定价字段白名单。
type userSupportedModelPricing struct {
BillingMode string `json:"billing_mode"`
InputPrice *float64 `json:"input_price"`
OutputPrice *float64 `json:"output_price"`
CacheWritePrice *float64 `json:"cache_write_price"`
CacheReadPrice *float64 `json:"cache_read_price"`
ImageInputPrice *float64 `json:"image_input_price"`
ImageOutputPrice *float64 `json:"image_output_price"`
PerRequestPrice *float64 `json:"per_request_price"`
Intervals []userPricingIntervalDTO `json:"intervals"`
BillingMode string `json:"billing_mode"`
InputPrice *float64 `json:"input_price"`
OutputPrice *float64 `json:"output_price"`
CacheWritePrice *float64 `json:"cache_write_price"`
CacheWrite1hPrice *float64 `json:"cache_write_1h_price"`
CacheReadPrice *float64 `json:"cache_read_price"`
ImageInputPrice *float64 `json:"image_input_price"`
ImageOutputPrice *float64 `json:"image_output_price"`
PerRequestPrice *float64 `json:"per_request_price"`
Intervals []userPricingIntervalDTO `json:"intervals"`
}
// userPricingIntervalDTO 定价区间白名单（去掉内部 ID、SortOrder 等前端不渲染的字段）。
type userPricingIntervalDTO struct {
MinTokens int `json:"min_tokens"`
MaxTokens *int `json:"max_tokens"`
TierLabel string `json:"tier_label,omitempty"`
InputPrice *float64 `json:"input_price"`
OutputPrice *float64 `json:"output_price"`
CacheWritePrice *float64 `json:"cache_write_price"`
CacheReadPrice *float64 `json:"cache_read_price"`
PerRequestPrice *float64 `json:"per_request_price"`
MinTokens int `json:"min_tokens"`
MaxTokens *int `json:"max_tokens"`
TierLabel string `json:"tier_label,omitempty"`
InputPrice *float64 `json:"input_price"`
OutputPrice *float64 `json:"output_price"`
CacheWritePrice *float64 `json:"cache_write_price"`
CacheWrite1hPrice *float64 `json:"cache_write_1h_price"`
CacheReadPrice *float64 `json:"cache_read_price"`
PerRequestPrice *float64 `json:"per_request_price"`
}
// userSupportedModel 用户可见的支持模型条目。
@@ -292,14 +294,15 @@ func toUserPricingIntervals(src []service.PricingInterval) []userPricingInterval
intervals := make([]userPricingIntervalDTO, 0, len(src))
for _, iv := range src {
intervals = append(intervals, userPricingIntervalDTO{
MinTokens: iv.MinTokens,
MaxTokens: iv.MaxTokens,
TierLabel: iv.TierLabel,
InputPrice: iv.InputPrice,
OutputPrice: iv.OutputPrice,
CacheWritePrice: iv.CacheWritePrice,
CacheReadPrice: iv.CacheReadPrice,
PerRequestPrice: iv.PerRequestPrice,
MinTokens: iv.MinTokens,
MaxTokens: iv.MaxTokens,
TierLabel: iv.TierLabel,
InputPrice: iv.InputPrice,
OutputPrice: iv.OutputPrice,
CacheWritePrice: iv.CacheWritePrice,
CacheWrite1hPrice: iv.CacheWrite1hPrice,
CacheReadPrice: iv.CacheReadPrice,
PerRequestPrice: iv.PerRequestPrice,
})
}
return intervals
@@ -320,14 +323,15 @@ func toUserPricing(p *service.ChannelModelPricing) *userSupportedModelPricing {
billingMode = string(service.BillingModeToken)
}
return &userSupportedModelPricing{
BillingMode: billingMode,
InputPrice: p.InputPrice,
OutputPrice: p.OutputPrice,
CacheWritePrice: p.CacheWritePrice,
CacheReadPrice: p.CacheReadPrice,
ImageInputPrice: p.ImageInputPrice,
ImageOutputPrice: p.ImageOutputPrice,
PerRequestPrice: p.PerRequestPrice,
Intervals: intervals,
BillingMode: billingMode,
InputPrice: p.InputPrice,
OutputPrice: p.OutputPrice,
CacheWritePrice: p.CacheWritePrice,
CacheWrite1hPrice: p.CacheWrite1hPrice,
CacheReadPrice: p.CacheReadPrice,
ImageInputPrice: p.ImageInputPrice,
ImageOutputPrice: p.ImageOutputPrice,
PerRequestPrice: p.PerRequestPrice,
Intervals: intervals,
}
}
backend/internal/handler/composite_platform.go
+20 -11
View File
@@ -1,6 +1,7 @@
package handler
import (
"net/http"
"strings"
"github.com/Wei-Shaw/sub2api/internal/service"
@@ -59,17 +60,17 @@ func effectiveAPIKeyPlatform(c *gin.Context, apiKey *service.APIKey) string {
return apiKey.Group.Platform
}
func openAIReasoningEffortPolicyForRequest(c *gin.Context, apiKey *service.APIKey) (string, []service.ReasoningEffortMapping, bool) {
func openAIReasoningEffortPolicyForRequest(c *gin.Context, apiKey *service.APIKey) (string, []service.ReasoningEffortMapping, string, bool) {
if apiKey == nil || apiKey.Group == nil {
return "", nil, false
return "", nil, "", false
}
if apiKey.Group.Platform != service.PlatformOpenAI && apiKey.Group.Platform != service.PlatformComposite {
return "", nil, false
return "", nil, "", false
}
if effectiveAPIKeyPlatform(c, apiKey) != service.PlatformOpenAI {
return "", nil, false
return "", nil, "", false
}
return apiKey.Group.MaxReasoningEffort, apiKey.Group.ReasoningEffortMappings, true
return apiKey.Group.MaxReasoningEffort, apiKey.Group.ReasoningEffortMappings, apiKey.Group.MaxReasoningEffortOverLimit, true
}
func bindRequestedReasoningEffort(c *gin.Context, body []byte, model string) {
@@ -100,13 +101,21 @@ func stampForwardRequestedReasoningEffort(result *service.ForwardResult, request
result.RequestedReasoningEffort = requested
}
func applyOpenAIReasoningEffortPolicyForRequest(c *gin.Context, apiKey *service.APIKey, body []byte) ([]byte, bool) {
func applyOpenAIReasoningEffortPolicyForRequest(c *gin.Context, apiKey *service.APIKey, body []byte) ([]byte, bool, error) {
bindRequestedReasoningEffort(c, body, strings.TrimSpace(gjson.GetBytes(body, "model").String()))
maxEffort, mappings, ok := openAIReasoningEffortPolicyForRequest(c, apiKey)
maxEffort, mappings, overLimit, ok := openAIReasoningEffortPolicyForRequest(c, apiKey)
if !ok {
return body, false
return body, false, nil
}
return service.ApplyOpenAIReasoningEffortPolicy(body, maxEffort, mappings)
return service.ApplyOpenAIReasoningEffortPolicy(body, maxEffort, mappings, overLimit)
}
func respondOpenAIReasoningEffortPolicyError(c *gin.Context, err error, write func(*gin.Context, int, string, string)) {
if c == nil || err == nil || write == nil {
return
}
service.MarkOpsClientBusinessLimited(c, service.OpsClientBusinessLimitedReasonLocalPolicyDenied)
write(c, http.StatusForbidden, "permission_error", err.Error())
}
func bindOpenAIReasoningEffortPolicyForMessagesRequest(c *gin.Context, apiKey *service.APIKey, body []byte) {
@@ -121,9 +130,9 @@ func bindOpenAIReasoningEffortPolicyForMessagesRequest(c *gin.Context, apiKey *s
if !effort.Exists() || effort.Type != gjson.String || strings.TrimSpace(effort.String()) == "" {
return
}
maxEffort, mappings, ok := openAIReasoningEffortPolicyForRequest(c, apiKey)
maxEffort, mappings, overLimit, ok := openAIReasoningEffortPolicyForRequest(c, apiKey)
if !ok {
return
}
c.Request = c.Request.WithContext(service.WithOpenAIReasoningEffortPolicy(c.Request.Context(), maxEffort, mappings))
c.Request = c.Request.WithContext(service.WithOpenAIReasoningEffortPolicy(c.Request.Context(), maxEffort, mappings, overLimit))
}
backend/internal/handler/composite_platform_test.go
+19 -4
View File
@@ -117,7 +117,8 @@ func TestOpenAIReasoningEffortPolicyForCompositeTarget(t *testing.T) {
openAICtx, _ := gin.CreateTestContext(httptest.NewRecorder())
openAICtx.Request = httptest.NewRequest("POST", "/v1/responses", nil)
openAICtx.Request = openAICtx.Request.WithContext(service.WithResolvedTargetPlatform(openAICtx.Request.Context(), service.PlatformOpenAI))
got, changed := applyOpenAIReasoningEffortPolicyForRequest(openAICtx, apiKey, body)
got, changed, err := applyOpenAIReasoningEffortPolicyForRequest(openAICtx, apiKey, body)
require.NoError(t, err)
require.True(t, changed)
require.JSONEq(t, `{"reasoning":{"effort":"medium"}}`, string(got))
requested := service.RequestedReasoningEffortFromContext(openAICtx.Request.Context())
@@ -125,7 +126,8 @@ func TestOpenAIReasoningEffortPolicyForCompositeTarget(t *testing.T) {
require.Equal(t, "max", *requested)
bindOpenAIReasoningEffortPolicyForMessagesRequest(openAICtx, apiKey, []byte(`{"output_config":{"effort":"max"}}`))
bound, changed := service.ApplyOpenAIReasoningEffortPolicyFromContext(openAICtx.Request.Context(), body)
bound, changed, err := service.ApplyOpenAIReasoningEffortPolicyFromContext(openAICtx.Request.Context(), body)
require.NoError(t, err)
require.True(t, changed)
require.JSONEq(t, `{"reasoning":{"effort":"medium"}}`, string(bound))
@@ -133,14 +135,27 @@ func TestOpenAIReasoningEffortPolicyForCompositeTarget(t *testing.T) {
omittedCtx.Request = httptest.NewRequest("POST", "/v1/messages", nil)
omittedCtx.Request = omittedCtx.Request.WithContext(service.WithResolvedTargetPlatform(omittedCtx.Request.Context(), service.PlatformOpenAI))
bindOpenAIReasoningEffortPolicyForMessagesRequest(omittedCtx, apiKey, []byte(`{"model":"gpt-5"}`))
omitted, changed := service.ApplyOpenAIReasoningEffortPolicyFromContext(omittedCtx.Request.Context(), body)
omitted, changed, err := service.ApplyOpenAIReasoningEffortPolicyFromContext(omittedCtx.Request.Context(), body)
require.NoError(t, err)
require.False(t, changed)
require.Equal(t, body, omitted)
denyGroup := *group
denyGroup.MaxReasoningEffortOverLimit = service.ReasoningEffortOverLimitDeny
denyAPIKey := &service.APIKey{Group: &denyGroup}
denyCtx, _ := gin.CreateTestContext(httptest.NewRecorder())
denyCtx.Request = httptest.NewRequest("POST", "/v1/responses", nil)
denyCtx.Request = denyCtx.Request.WithContext(service.WithResolvedTargetPlatform(denyCtx.Request.Context(), service.PlatformOpenAI))
_, _, err = applyOpenAIReasoningEffortPolicyForRequest(denyCtx, denyAPIKey, body)
require.Error(t, err)
var overLimit *service.ReasoningEffortOverLimitError
require.ErrorAs(t, err, &overLimit)
grokCtx, _ := gin.CreateTestContext(httptest.NewRecorder())
grokCtx.Request = httptest.NewRequest("POST", "/v1/responses", nil)
grokCtx.Request = grokCtx.Request.WithContext(service.WithResolvedTargetPlatform(grokCtx.Request.Context(), service.PlatformGrok))
got, changed = applyOpenAIReasoningEffortPolicyForRequest(grokCtx, apiKey, body)
got, changed, err = applyOpenAIReasoningEffortPolicyForRequest(grokCtx, apiKey, body)
require.NoError(t, err)
require.False(t, changed)
require.Equal(t, body, got)
}
backend/internal/handler/dto/group_mapper_openai_fast_test.go
+25
View File
@@ -0,0 +1,25 @@
package dto
import (
"encoding/json"
"testing"
"github.com/Wei-Shaw/sub2api/internal/service"
"github.com/stretchr/testify/require"
)
func TestGroupMapperExposesForceOpenAIFastOnlyToAdmins(t *testing.T) {
group := &service.Group{
ID: 7, Name: "fast", Platform: service.PlatformOpenAI, Status: service.StatusActive, ForceOpenAIFast: true, FreeOpenAIFast: true,
}
userJSON, err := json.Marshal(GroupFromService(group))
require.NoError(t, err)
require.NotContains(t, string(userJSON), "force_openai_fast")
require.NotContains(t, string(userJSON), "free_openai_fast")
adminJSON, err := json.Marshal(GroupFromServiceAdmin(group))
require.NoError(t, err)
require.Contains(t, string(adminJSON), `"force_openai_fast":true`)
require.Contains(t, string(adminJSON), `"free_openai_fast":true`)
}
backend/internal/handler/dto/mappers.go
+3
View File
@@ -148,6 +148,8 @@ func GroupFromServiceAdmin(g *service.Group) *AdminGroup {
}
out := &AdminGroup{
Group: groupFromServiceBase(g),
ForceOpenAIFast: g.ForceOpenAIFast,
FreeOpenAIFast: g.FreeOpenAIFast,
ProfitControlEnabled: g.ProfitControlEnabled,
ProfitMinMargin: g.ProfitMinMargin,
ProfitSafetyBuffer: g.ProfitSafetyBuffer,
@@ -221,6 +223,7 @@ func groupFromServiceBase(g *service.Group) Group {
RequirePrivacySet: g.RequirePrivacySet,
RPMLimit: g.RPMLimit,
MaxReasoningEffort: g.MaxReasoningEffort,
MaxReasoningEffortOverLimit: g.MaxReasoningEffortOverLimit,
ReasoningEffortMappings: g.ReasoningEffortMappings,
CreatedAt: g.CreatedAt,
UpdatedAt: g.UpdatedAt,
backend/internal/handler/dto/types.go
+6
View File
@@ -153,6 +153,8 @@ type Group struct {
RPMLimit int `json:"rpm_limit"`
// MaxReasoningEffort OpenAI/Codex 请求的推理强度上限，空字符串表示不限制。
MaxReasoningEffort string `json:"max_reasoning_effort"`
// MaxReasoningEffortOverLimit 超过上限时的访问控制：downgrade（默认）或 deny。
MaxReasoningEffortOverLimit string `json:"max_reasoning_effort_over_limit"`
// ReasoningEffortMappings OpenAI/Codex 推理强度映射，可按模型精确名、前缀或后缀限定。
ReasoningEffortMappings []domain.ReasoningEffortMapping `json:"reasoning_effort_mappings"`
@@ -164,6 +166,10 @@ type Group struct {
// 注意：普通用户接口不得返回 model_routing/account_count/account_groups 等内部信息。
type AdminGroup struct {
Group
// ForceOpenAIFast 是管理端请求策略，用户侧分组 DTO 无需暴露。
ForceOpenAIFast bool `json:"force_openai_fast"`
// FreeOpenAIFast 是管理端计费策略，用户侧分组 DTO 无需暴露。
FreeOpenAIFast bool `json:"free_openai_fast"`
// 分组利润控制（五个 token 平台分组可启用；margin/buffer 为小数存储）。
// 仅管理员可见：这三个字段与同响应中的 rate_multiplier 相乘即可反推出
backend/internal/handler/endpoint.go
+1 -1
View File
@@ -279,7 +279,7 @@ func InboundEndpointMiddleware() gin.HandlerFunc {
// ──────────────────────────────────────────────────────────
// Context helpers — used by handlers before building
// RecordUsageInput / RecordUsageLongContextInput.
// RecordUsageInput.
// ──────────────────────────────────────────────────────────
// GetInboundEndpoint returns the canonical inbound endpoint stored by
backend/internal/handler/gateway_models_test.go
+7
View File
@@ -72,6 +72,13 @@ func (s *gatewayModelsAccountRepoStub) ListByGroup(ctx context.Context, groupID
return s.ListSchedulableByGroupID(ctx, groupID)
}
func (s *gatewayModelsAccountRepoStub) ListModelAvailabilityCandidates(ctx context.Context, groupID *int64, _ []string, _ bool) ([]service.Account, error) {
if groupID == nil {
return nil, nil
}
return s.ListSchedulableByGroupID(ctx, *groupID)
}
func newGatewayModelsHandlerForTest(repo service.AccountRepository) *GatewayHandler {
return &GatewayHandler{
gatewayService: service.NewGatewayService(
backend/internal/handler/gemini_v1beta_handler.go
+18 -26
View File
@@ -569,33 +569,25 @@ func (h *GatewayHandler) GeminiV1BetaModels(c *gin.Context) {
forceCacheBilling := fs.ForceCacheBilling
quotaPlatform := service.QuotaPlatform(c.Request.Context(), apiKey)
sessionID := service.ExtractClientSessionID(c)
// 长上下文规则由计费服务统一持有（模型广场展示同源），入口只负责声明自己适用该规则。
var longContextThreshold int
var longContextMultiplier float64
if rule := h.gatewayService.LegacyLongContextRule(service.PlatformGemini); rule != nil {
longContextThreshold = rule.Threshold
longContextMultiplier = rule.Multiplier
}
// 长上下文阶梯由目录数据驱动，统一在计费路径内生效，入口无需声明。
h.submitUsageRecordTask(c.Request.Context(), func(ctx context.Context) {
if err := h.gatewayService.RecordUsageWithLongContext(ctx, &service.RecordUsageLongContextInput{
Result: result,
QuotaPlatform: quotaPlatform,
APIKey: apiKey,
User: apiKey.User,
Account: account,
Subscription: subscription,
PricingAt: pricingAt,
InboundEndpoint: inboundEndpoint,
UpstreamEndpoint: upstreamEndpoint,
UserAgent: userAgent,
IPAddress: clientIP,
RequestPayloadHash: requestPayloadHash,
LongContextThreshold: longContextThreshold,
LongContextMultiplier: longContextMultiplier,
ForceCacheBilling: forceCacheBilling,
APIKeyService: h.apiKeyService,
SessionID: sessionID,
ChannelUsageFields: clientRequestedUsageFields(c, channelMapping, reqModel, result.UpstreamModel),
if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{
Result: result,
QuotaPlatform: quotaPlatform,
APIKey: apiKey,
User: apiKey.User,
Account: account,
Subscription: subscription,
PricingAt: pricingAt,
InboundEndpoint: inboundEndpoint,
UpstreamEndpoint: upstreamEndpoint,
UserAgent: userAgent,
IPAddress: clientIP,
RequestPayloadHash: requestPayloadHash,
ForceCacheBilling: forceCacheBilling,
APIKeyService: h.apiKeyService,
SessionID: sessionID,
ChannelUsageFields: clientRequestedUsageFields(c, channelMapping, reqModel, result.UpstreamModel),
}); err != nil {
logger.L().With(
zap.String("component", "handler.gemini_v1beta.models"),
backend/internal/handler/no_account_error.go
+17
View File
@@ -42,6 +42,23 @@ func classifySelectionFailureError(err error, fallback noAccountErrorClassificat
if err == nil {
return fallback
}
// A 404 model_not_found fallback is authoritative and must not be downgraded
// to a rate-limit verdict. classifyNoAccountError only reaches it through
// DiagnoseModelAvailabilityForPlatform, a dedicated database query over
// persistent eligibility (active + schedulable + model_mapping) that already
// established no account in the group can serve this model at all. A transient
// per-model cooldown on one of the remaining candidates does not make "all
// available accounts are rate-limited" true.
//
// Reporting 429 here is actively harmful: retrying can never succeed, and
// clients that treat 429 as a rate limit retry hard and swallow the body
// (Codex surfaces only "exceeded retry limit, last status: 429"), losing the
// one message that names the real problem. It also flips the ops attribution
// from a local model-configuration issue to routing capacity, because call
// sites gate markOpsRoutingCapacityLimitedIfNoAvailable on ModelNotFound.
if fallback.ModelNotFound {
return fallback
}
match := selectionModelRateLimitedPattern.FindStringSubmatch(strings.ToLower(err.Error()))
if len(match) != 2 {
return fallback
backend/internal/handler/no_account_error_test.go
+69
View File
@@ -234,3 +234,72 @@ func TestClassifyNoAccountError_FromGin_NilContextStillSafe(t *testing.T) {
require.False(t, service.HasOpsClientBusinessLimited(nil))
require.Empty(t, service.OpsClientBusinessLimitedReason(nil))
}
// 权威的 404 model_not_found 不能被"账号被限流"的 429 盖掉。
//
// 选号失败的错误串同时携带多种过滤原因，例如
// "pool=9, filtered: model_not_supported=8 model_rate_limited=1"：8 个账号根本不支持该模型，
// 剩下 1 个恰好处于模型级冷却。此时 classifyNoAccountError 已通过持久化判据确认整个分组
// 没有账号能服务该模型（ModelNotFound=true），改判成 429 "All available accounts are
// currently rate-limited" 是错误诊断——重试永远不会成功，而把 429 当限流的客户端会反复
// 重试并吞掉 body（Codex 只显示 "exceeded retry limit"），恰好丢掉唯一说明真实原因的信息。
func TestClassifySelectionFailureError_ModelNotFoundIsNotOverriddenByRateLimited(t *testing.T) {
modelNotFound := noAccountErrorClassification{
Status: http.StatusNotFound,
ErrType: "model_not_found",
Message: `Model "gpt-5.3-codex" is not supported by any configured account in this group`,
ModelNotFound: true,
}
got := classifySelectionFailureError(
fmt.Errorf("no available OpenAI accounts supporting model: gpt-5.3-codex "+
"(pool=9, filtered: model_not_supported=8 model_rate_limited=1)"),
modelNotFound,
)
require.Equal(t, modelNotFound, got,
"分组里没有任何账号能服务该模型时，模型级冷却不该把 404 改判成 429")
}
// 真实调用点的顺序：先 classifyNoAccountErrorFromGin，再 classifySelectionFailureError。
// 覆盖这条链路是为了同时锁住 ops 归因——调用点用 ModelNotFound 决定是否标记
// routing capacity limited，一旦 404 被改判成 429，同一个请求会既被标成
// local model configuration 又被标成容量问题，自相矛盾。
func TestClassifySelectionFailureError_CallSiteChainKeepsModelNotFoundAttribution(t *testing.T) {
c := newTestGinContextWithRequest()
fd := &fakeDiagnoser{resp: service.ModelAvailabilityDiagnosis{HasAccountsInPool: true, HasModelSupport: false}}
apiKey := &service.APIKey{GroupID: ptrInt64(43)}
cls := classifyNoAccountErrorFromGin(c, fd, apiKey, "gpt-5.3-codex", "gpt-5.3-codex", service.PlatformOpenAI)
cls = classifySelectionFailureError(
fmt.Errorf("no available OpenAI accounts supporting model: gpt-5.3-codex "+
"(pool=9, filtered: model_not_supported=8 model_rate_limited=1)"),
cls,
)
require.Equal(t, http.StatusNotFound, cls.Status)
require.Equal(t, "model_not_found", cls.ErrType)
require.True(t, cls.ModelNotFound)
require.Contains(t, cls.Message, "gpt-5.3-codex")
require.True(t, service.HasOpsClientBusinessLimited(c))
require.Equal(t, service.OpsClientBusinessLimitedReasonLocalModelConfiguration, service.OpsClientBusinessLimitedReason(c))
}
// 池子里确实存在能服务该模型、只是全部在冷却的账号时，429 改判仍需保留：
// 这种情况 fallback 是 503（HasModelSupport=true），重试是有意义的。
func TestClassifySelectionFailureError_StillUpgradesNonModelNotFoundFallback(t *testing.T) {
fallback := noAccountErrorClassification{
Status: http.StatusServiceUnavailable,
ErrType: "api_error",
Message: "Service temporarily unavailable",
}
got := classifySelectionFailureError(
fmt.Errorf("no available accounts supporting model: gpt-5.6-sol (total=3 eligible=0 model_rate_limited=3)"),
fallback,
)
require.Equal(t, http.StatusTooManyRequests, got.Status)
require.Equal(t, "rate_limit_error", got.ErrType)
require.False(t, got.ModelNotFound)
}
backend/internal/handler/openai_automation_bootstrap_test.go
+133
View File
@@ -0,0 +1,133 @@
package handler
import (
"strings"
"testing"
"github.com/stretchr/testify/require"
"github.com/tidwall/gjson"
)
const automationBootstrapPrompt = "Review the project and report any important changes."
func codexAutomationBootstrap(automationID, lastRun, prompt string) string {
return "Automation: Scheduled project review\n" +
"Automation ID: " + automationID + "\n" +
"Automation memory: $CODEX_HOME/automations/" + automationID + "/memory.md\n" +
"Last run: " + lastRun + "\n\n" + prompt
}
func codexAutomationBootstrapBody(t *testing.T, output, callID string) []byte {
t.Helper()
return []byte(`{"model":"gpt-5","input":[{"type":"function_call_output","namespace":"codex_app","name":"automation_update","output":` +
mustJSON(t, output) + callID + `}]}`)
}
func TestNormalizeCodexAutomationBootstrapSupportedLastRunValues(t *testing.T) {
tests := []struct {
name string
lastRun string
crlf bool
}{
{name: "never", lastRun: "never"},
{name: "timestamp", lastRun: "2026-09-01T02:06:34.536Z (1788228394536)"},
{name: "crlf", lastRun: "never", crlf: true},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
output := codexAutomationBootstrap("wiki-maintenance", tt.lastRun, automationBootstrapPrompt)
if tt.crlf {
output = strings.ReplaceAll(output, "\n", "\r\n")
}
got, changed := normalizeCodexAutomationBootstrap(codexAutomationBootstrapBody(t, output, ""))
require.True(t, changed)
require.Equal(t, "message", gjson.GetBytes(got, "input.0.type").String())
require.Equal(t, "user", gjson.GetBytes(got, "input.0.role").String())
require.Equal(t, output, gjson.GetBytes(got, "input.0.content.0.text").String())
require.False(t, gjson.GetBytes(got, "input.0.call_id").Exists())
})
}
}
func TestNormalizeCodexAutomationBootstrapRejectsUnsafeShapes(t *testing.T) {
validOutput := codexAutomationBootstrap("wiki", "never", automationBootstrapPrompt)
tests := []struct {
name string
body []byte
}{
{
name: "ordinary missing call id",
body: []byte(`{"model":"gpt-5","input":[{"type":"function_call_output","namespace":"codex_app","name":"other","output":` + mustJSON(t, validOutput) + `}]}`),
},
{
name: "tui namespace",
body: []byte(`{"model":"gpt-5","input":[{"type":"function_call_output","namespace":"codex_tui","name":"automation_update","output":` + mustJSON(t, validOutput) + `}]}`),
},
{
name: "valid call id",
body: codexAutomationBootstrapBody(t, validOutput, `,"call_id":"call-1"`),
},
{
name: "previous response",
body: []byte(`{"model":"gpt-5","previous_response_id":"resp-1","input":[{"type":"function_call_output","namespace":"codex_app","name":"automation_update","output":` + mustJSON(t, validOutput) + `}]}`),
},
{
name: "real call context",
body: []byte(`{"model":"gpt-5","input":[{"type":"function_call_output","namespace":"codex_app","name":"automation_update","output":` + mustJSON(t, validOutput) + `},{"type":"function_call","call_id":"call-1"}]}`),
},
{
name: "heartbeat output",
body: codexAutomationBootstrapBody(t, `<heartbeat><automation_id>wiki</automation_id></heartbeat>`, ""),
},
{
name: "mismatched memory id",
body: codexAutomationBootstrapBody(t, strings.Replace(validOutput, "/wiki/memory.md", "/other/memory.md", 1), ""),
},
{
name: "unsafe automation id",
body: codexAutomationBootstrapBody(t, codexAutomationBootstrap("../wiki", "never", automationBootstrapPrompt), ""),
},
{
name: "invalid timestamp",
body: codexAutomationBootstrapBody(t, codexAutomationBootstrap("wiki", "yesterday", automationBootstrapPrompt), ""),
},
{
name: "mismatched timestamp epoch",
body: codexAutomationBootstrapBody(t, codexAutomationBootstrap("wiki", "2026-09-01T02:06:34.536Z (1)", automationBootstrapPrompt), ""),
},
{
name: "missing separator",
body: codexAutomationBootstrapBody(t, strings.Replace(validOutput, "\n\n"+automationBootstrapPrompt, "\n"+automationBootstrapPrompt, 1), ""),
},
{
name: "empty prompt",
body: codexAutomationBootstrapBody(t, codexAutomationBootstrap("wiki", "never", " \n"), ""),
},
{
name: "mixed missing call id output",
body: []byte(`{"model":"gpt-5","input":[{"type":"function_call_output","namespace":"codex_app","name":"automation_update","output":` + mustJSON(t, validOutput) + `},{"type":"computer_call_output","output":"done"}]}`),
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
got, changed := normalizeCodexAutomationBootstrap(tt.body)
require.False(t, changed)
require.Equal(t, tt.body, got)
})
}
}
func TestNormalizeCodexAutomationBootstrapPreservesOrderAndIsIdempotent(t *testing.T) {
output := codexAutomationBootstrap("wiki", "never", automationBootstrapPrompt)
body := []byte(`{"model":"gpt-5","input":[{"type":"message","role":"user","content":"before"},{"type":"function_call_output","namespace":"codex_app","name":"automation_update","output":` + mustJSON(t, output) + `},{"type":"message","role":"user","content":"after"}]}`)
got, changed := normalizeCodexAutomationBootstrap(body)
require.True(t, changed)
require.Equal(t, "before", gjson.GetBytes(got, "input.0.content").String())
require.Equal(t, output, gjson.GetBytes(got, "input.1.content.0.text").String())
require.Equal(t, "after", gjson.GetBytes(got, "input.2.content").String())
again, changedAgain := normalizeCodexAutomationBootstrap(got)
require.False(t, changedAgain)
require.Equal(t, got, again)
}
backend/internal/handler/openai_chat_completions.go
+4 -1
View File
@@ -81,7 +81,10 @@ func (h *OpenAIGatewayHandler) ChatCompletions(c *gin.Context) {
h.errorResponse(c, http.StatusBadRequest, "invalid_request_error", "Model is not supported by this OpenAI-compatible endpoint for composite groups")
return
}
if cappedBody, changed := applyOpenAIReasoningEffortPolicyForRequest(c, apiKey, body); changed {
if cappedBody, changed, err := applyOpenAIReasoningEffortPolicyForRequest(c, apiKey, body); err != nil {
respondOpenAIReasoningEffortPolicyError(c, err, h.errorResponse)
return
} else if changed {
body = cappedBody
}
reqStream, ok := parseOpenAICompatibleStream(body)
backend/internal/handler/openai_codex_models_handler_test.go
+6 -2
View File
@@ -53,6 +53,10 @@ func (r codexModelsFailoverAccountRepo) ListByGroup(_ context.Context, _ int64)
return append([]service.Account(nil), r.accounts...), nil
}
func (r codexModelsFailoverAccountRepo) ListModelAvailabilityCandidates(_ context.Context, _ *int64, _ []string, _ bool) ([]service.Account, error) {
return append([]service.Account(nil), r.accounts...), nil
}
type codexModelsFailoverHTTPUpstream struct {
service.HTTPUpstream
mu sync.Mutex
@@ -620,8 +624,8 @@ func requireCompleteCodexModelsHandlerResponse(t *testing.T, recorder *httptest.
t.Fatalf("truncation_policy must be populated: %v", model["truncation_policy"])
}
modalities, ok := model["input_modalities"].([]any)
if !ok || len(modalities) != 1 || modalities[0] != "text" {
t.Fatalf("custom OpenAI-compatible endpoint modalities: got %v, want [text]", model["input_modalities"])
if !ok || len(modalities) != 2 || modalities[0] != "text" || modalities[1] != "image" {
t.Fatalf("known GPT model modalities: got %v, want [text image]", model["input_modalities"])
}
}
backend/internal/handler/openai_delegation_bootstrap_test.go
+176
View File
@@ -0,0 +1,176 @@
package handler
import (
"encoding/json"
"testing"
"github.com/stretchr/testify/require"
"github.com/tidwall/gjson"
)
const delegationEnvelope = `<codex_delegation><source_thread_id>thread-1</source_thread_id><input>do the work</input></codex_delegation>`
func TestNormalizeCodexDelegationBootstrapSupportedTools(t *testing.T) {
tests := []struct {
name string
namespace string
tool string
callID string
}{
{name: "app create missing call id", namespace: "codex_app", tool: "create_thread"},
{name: "app send empty call id", namespace: "codex_app", tool: "send_message_to_thread", callID: `,"call_id":""`},
{name: "tui create whitespace call id", namespace: "codex_tui", tool: "create_thread", callID: `,"call_id":" "`},
{name: "tui send missing call id", namespace: "codex_tui", tool: "send_message_to_thread"},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
body := []byte(`{"model":"gpt-5","input":[{"type":"function_call_output","namespace":"` + tt.namespace + `","name":"` + tt.tool + `","output":"` + delegationEnvelope + `"` + tt.callID + `}]}`)
got, changed := normalizeCodexDelegationBootstrap(body)
require.True(t, changed)
require.Equal(t, "message", gjson.GetBytes(got, "input.0.type").String())
require.Equal(t, "user", gjson.GetBytes(got, "input.0.role").String())
require.Equal(t, delegationEnvelope, gjson.GetBytes(got, "input.0.content.0.text").String())
require.False(t, gjson.GetBytes(got, "input.0.call_id").Exists())
})
}
}
func TestNormalizeCodexDelegationBootstrapRejectsUnsafeShapes(t *testing.T) {
tests := []struct {
name string
item string
extra string
}{
{name: "ordinary missing call id", item: `{"type":"function_call_output","name":"other","namespace":"codex_app","output":"` + delegationEnvelope + `"}`},
{name: "valid call id", item: `{"type":"function_call_output","call_id":"call-1","name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"}`},
{name: "previous response", item: `{"type":"function_call_output","name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"}`, extra: `,"previous_response_id":"resp-1"`},
{name: "real call context", item: `{"type":"function_call_output","name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"},{"type":"function_call","call_id":"call-1"}`},
{name: "ambiguous call context", item: `{"type":"function_call_output","name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"},{"type":"function_call"}`},
{name: "item reference context", item: `{"type":"function_call_output","name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"},{"type":"item_reference","id":"call-1"}`},
{name: "built-in call context", item: `{"type":"function_call_output","name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"},{"type":"computer_call","call_id":"call-1"}`},
{name: "ambiguous built-in call context", item: `{"type":"function_call_output","name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"},{"type":"computer_call"}`},
{name: "mixed missing call id output", item: `{"type":"function_call_output","name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"},{"type":"computer_call_output","output":"done"}`},
{name: "mixed tool search output", item: `{"type":"function_call_output","name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"},{"type":"tool_search_output","output":"done"}`},
{name: "mixed invalid delegation output", item: `{"type":"function_call_output","name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"},{"type":"function_call_output","name":"create_thread","namespace":"codex_app","output":"not an envelope"}`},
{name: "non string call id", item: `{"type":"function_call_output","call_id":42,"name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"}`},
{name: "padded type", item: `{"type":" function_call_output ","name":"create_thread","namespace":"codex_app","output":"` + delegationEnvelope + `"}`},
{name: "padded name", item: `{"type":"function_call_output","name":" create_thread ","namespace":"codex_app","output":"` + delegationEnvelope + `"}`},
{name: "padded namespace", item: `{"type":"function_call_output","name":"create_thread","namespace":" codex_app ","output":"` + delegationEnvelope + `"}`},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
body := []byte(`{"model":"gpt-5","input":[` + tt.item + `]` + tt.extra + `}`)
got, changed := normalizeCodexDelegationBootstrap(body)
require.False(t, changed)
require.Equal(t, body, got)
})
}
}
func TestNormalizeCodexDelegationBootstrapPreviousResponseID(t *testing.T) {
tests := []struct {
name string
value string
changed bool
}{
{name: "absent", changed: true},
{name: "blank string", value: `,"previous_response_id":" "`, changed: true},
{name: "non-empty string", value: `,"previous_response_id":"resp-1"`},
{name: "null", value: `,"previous_response_id":null`},
{name: "number", value: `,"previous_response_id":42`},
{name: "boolean", value: `,"previous_response_id":false`},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
body := []byte(`{"model":"gpt-5","input":[{"type":"function_call_output","namespace":"codex_app","name":"create_thread","output":"` + delegationEnvelope + `"}]` + tt.value + `}`)
got, changed := normalizeCodexDelegationBootstrap(body)
require.Equal(t, tt.changed, changed)
if !tt.changed {
require.Equal(t, body, got)
}
})
}
}
func TestNormalizeCodexDelegationBootstrapRejectsDuplicateMembers(t *testing.T) {
tests := []struct {
name string
body string
}{
{
name: "top-level previous response id",
body: `{"model":"gpt-5","previous_response_id":"","previous_response_id":"resp-1","input":[{"type":"function_call_output","namespace":"codex_app","name":"create_thread","output":"` + delegationEnvelope + `"}]}`,
},
{
name: "candidate discriminator",
body: `{"model":"gpt-5","input":[{"type":"message","type":"function_call_output","namespace":"codex_app","name":"create_thread","output":"` + delegationEnvelope + `"}]}`,
},
{
name: "candidate call id",
body: `{"model":"gpt-5","input":[{"type":"function_call_output","namespace":"codex_app","name":"create_thread","call_id":"call-1","call_id":"","output":"` + delegationEnvelope + `"}]}`,
},
{
name: "candidate output",
body: `{"model":"gpt-5","input":[{"type":"function_call_output","namespace":"codex_app","name":"create_thread","output":"not a delegation","output":"` + delegationEnvelope + `"}]}`,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
body := []byte(tt.body)
got, changed := normalizeCodexDelegationBootstrap(body)
require.False(t, changed)
require.Equal(t, body, got)
})
}
}
func TestNormalizeCodexDelegationBootstrapPreservesExactNumbers(t *testing.T) {
body := []byte(`{"model":"gpt-5","metadata":{"integer":9007199254740993},"input":[{"type":"function_call_output","namespace":"codex_app","name":"create_thread","output":"` + delegationEnvelope + `"}]}`)
got, changed := normalizeCodexDelegationBootstrap(body)
require.True(t, changed)
require.Equal(t, "9007199254740993", gjson.GetBytes(got, "metadata.integer").Raw)
}
func TestNormalizeCodexDelegationBootstrapRequiresCompleteEnvelope(t *testing.T) {
malformed := []string{
`<codex_delegation><source_thread_id>thread-1</source_thread_id></codex_delegation>`,
`<codex_delegation><source_thread_id></source_thread_id><input>work</input></codex_delegation>`,
`<codex_delegation><source_thread_id>thread-1</source_thread_id><input> </input></codex_delegation>`,
`prefix<codex_delegation><source_thread_id>thread-1</source_thread_id><input>work</input></codex_delegation>`,
`<codex_delegation><source_thread_id>thread-1</source_thread_id><input>work</input></codex_delegation>suffix`,
`<codex_delegation><source_thread_id>thread-1</source_thread_id><input>work</input><extra>x</extra></codex_delegation>`,
`<codex_delegation><source_thread_id>thread-1</source_thread_id><input><nested>work</nested></input></codex_delegation>`,
`<x:codex_delegation xmlns:x="urn:test"><source_thread_id>thread-1</source_thread_id><input>work</input></x:codex_delegation>`,
`<x:codex_delegation><source_thread_id>thread-1</source_thread_id><input>work</input></x:codex_delegation>`,
`<codex_delegation xmlns="urn:test"><source_thread_id>thread-1</source_thread_id><input>work</input></codex_delegation>`,
`<codex_delegation xmlns:x="urn:test"><x:source_thread_id>thread-1</x:source_thread_id><input>work</input></codex_delegation>`,
`<codex_delegation><x:source_thread_id>thread-1</x:source_thread_id><input>work</input></codex_delegation>`,
`<codex_delegation xmlns=""><source_thread_id>thread-1</source_thread_id><input>work</input></codex_delegation>`,
}
for _, envelope := range malformed {
body := []byte(`{"model":"gpt-5","input":[{"type":"function_call_output","namespace":"codex_app","name":"create_thread","output":` + mustJSON(t, envelope) + `}]}`)
got, changed := normalizeCodexDelegationBootstrap(body)
require.False(t, changed, envelope)
require.Equal(t, body, got)
}
}
func TestNormalizeCodexDelegationBootstrapPreservesOrderAndIsIdempotent(t *testing.T) {
body := []byte(`{"model":"gpt-5","input":[{"type":"message","role":"user","content":"before"},{"type":"function_call_output","namespace":"codex_tui","name":"send_message_to_thread","output":"` + delegationEnvelope + `"},{"type":"message","role":"user","content":"after"}]}`)
got, changed := normalizeCodexDelegationBootstrap(body)
require.True(t, changed)
require.Equal(t, "before", gjson.GetBytes(got, "input.0.content").String())
require.Equal(t, delegationEnvelope, gjson.GetBytes(got, "input.1.content.0.text").String())
require.Equal(t, "after", gjson.GetBytes(got, "input.2.content").String())
again, changedAgain := normalizeCodexDelegationBootstrap(got)
require.False(t, changedAgain)
require.Equal(t, got, again)
}
func mustJSON(t *testing.T, value string) string {
t.Helper()
encoded, err := json.Marshal(value)
require.NoError(t, err)
return string(encoded)
}
backend/internal/handler/openai_gateway_handler.go
+333 -8
View File
@@ -1,10 +1,13 @@
package handler
import (
"bytes"
"context"
"encoding/json"
"encoding/xml"
"errors"
"fmt"
"io"
"net/http"
"runtime/debug"
"strconv"
@@ -432,9 +435,24 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
h.errorResponse(c, http.StatusBadRequest, "invalid_request_error", "Model is not supported by this OpenAI-compatible endpoint for composite groups")
return
}
if cappedBody, changed := applyOpenAIReasoningEffortPolicyForRequest(c, apiKey, body); changed {
if cappedBody, changed, err := applyOpenAIReasoningEffortPolicyForRequest(c, apiKey, body); err != nil {
respondOpenAIReasoningEffortPolicyError(c, err, h.errorResponse)
return
} else if changed {
body = cappedBody
}
if normalizedBody, changed := normalizeCodexAutomationBootstrap(body); changed {
body = normalizedBody
reqLog.Info("openai.codex_automation_bootstrap_normalized",
zap.String("normalization", "call_output_to_user_message"),
)
}
if normalizedBody, changed := normalizeCodexDelegationBootstrap(body); changed {
body = normalizedBody
reqLog.Info("openai.codex_delegation_bootstrap_normalized",
zap.String("normalization", "call_output_to_user_message"),
)
}
reqStream, ok := parseOpenAICompatibleStream(body)
if !ok {
@@ -1581,6 +1599,312 @@ func (h *OpenAIGatewayHandler) validateFunctionCallOutputRequest(c *gin.Context,
return false
}
func normalizeCodexDelegationBootstrap(body []byte) ([]byte, bool) {
return normalizeCodexCallOutputBootstrap(body, isCodexDelegationCandidate)
}
func normalizeCodexAutomationBootstrap(body []byte) ([]byte, bool) {
return normalizeCodexCallOutputBootstrap(body, isCodexAutomationCandidate)
}
func normalizeCodexCallOutputBootstrap(body []byte, isCandidate func(map[string]any) bool) ([]byte, bool) {
if !hasUniqueJSONMembers(body) {
return body, false
}
decoder := json.NewDecoder(bytes.NewReader(body))
decoder.UseNumber()
var request map[string]any
if err := decoder.Decode(&request); err != nil {
return body, false
}
if previousResponseID, exists := request["previous_response_id"]; exists {
value, ok := previousResponseID.(string)
if !ok || strings.TrimSpace(value) != "" {
return body, false
}
}
input, ok := request["input"].([]any)
if !ok {
return body, false
}
// Any call/reference anchor makes a call-less output ambiguous. Responses
// built-ins follow the *_call / *_call_output naming convention, so classify
// by the wire type shape instead of maintaining an incomplete allowlist.
for _, raw := range input {
item, ok := raw.(map[string]any)
if !ok {
continue
}
typ := stringField(item, "type")
if typ == "item_reference" || strings.HasSuffix(typ, "_call") {
return body, false
}
if isResponsesCallOutputType(typ) {
callIDValue, exists := item["call_id"]
callID, isString := callIDValue.(string)
if exists && (!isString || strings.TrimSpace(callID) != "") {
return body, false
}
if !isCandidate(item) {
return body, false
}
}
}
changed := false
for i, raw := range input {
item, ok := raw.(map[string]any)
if !ok || !isCandidate(item) {
continue
}
output, ok := item["output"].(string)
if !ok {
continue
}
input[i] = map[string]any{
"type": "message",
"role": "user",
"content": []any{map[string]any{
"type": "input_text",
"text": output,
}},
}
changed = true
}
if !changed {
return body, false
}
normalized, err := json.Marshal(request)
if err != nil {
return body, false
}
return normalized, true
}
func hasUniqueJSONMembers(body []byte) bool {
decoder := json.NewDecoder(bytes.NewReader(body))
if !consumeUniqueJSONValue(decoder) {
return false
}
_, err := decoder.Token()
return err == io.EOF
}
func consumeUniqueJSONValue(decoder *json.Decoder) bool {
token, err := decoder.Token()
if err != nil {
return false
}
delim, ok := token.(json.Delim)
if !ok {
return true
}
switch delim {
case '{':
members := make(map[string]struct{})
for decoder.More() {
keyToken, err := decoder.Token()
if err != nil {
return false
}
key, ok := keyToken.(string)
if !ok {
return false
}
if _, duplicate := members[key]; duplicate {
return false
}
members[key] = struct{}{}
if !consumeUniqueJSONValue(decoder) {
return false
}
}
end, err := decoder.Token()
return err == nil && end == json.Delim('}')
case '[':
for decoder.More() {
if !consumeUniqueJSONValue(decoder) {
return false
}
}
end, err := decoder.Token()
return err == nil && end == json.Delim(']')
default:
return false
}
}
func isResponsesCallOutputType(typ string) bool {
return strings.HasSuffix(typ, "_call_output") || typ == "tool_search_output"
}
func isCodexDelegationCandidate(item map[string]any) bool {
if stringField(item, "type") != "function_call_output" ||
!isCodexDelegationTool(stringField(item, "namespace"), stringField(item, "name")) {
return false
}
output, ok := item["output"].(string)
return ok && validCodexDelegationEnvelope(output)
}
func isCodexAutomationCandidate(item map[string]any) bool {
if stringField(item, "type") != "function_call_output" ||
stringField(item, "namespace") != "codex_app" ||
stringField(item, "name") != "automation_update" {
return false
}
output, ok := item["output"].(string)
return ok && validCodexAutomationBootstrap(output)
}
func stringField(item map[string]any, key string) string {
value, _ := item[key].(string)
return value
}
func isCodexDelegationTool(namespace, name string) bool {
return (namespace == "codex_app" || namespace == "codex_tui") &&
(name == "create_thread" || name == "send_message_to_thread")
}
func validCodexAutomationBootstrap(value string) bool {
normalized := strings.ReplaceAll(value, "\r\n", "\n")
if strings.ContainsRune(normalized, '\r') {
return false
}
lines := strings.Split(normalized, "\n")
if len(lines) < 6 {
return false
}
if _, ok := codexAutomationHeaderValue(lines[0], "Automation: "); !ok {
return false
}
automationID, ok := codexAutomationHeaderValue(lines[1], "Automation ID: ")
if !ok || !validCodexAutomationID(automationID) {
return false
}
expectedMemory := "Automation memory: $CODEX_HOME/automations/" + automationID + "/memory.md"
if lines[2] != expectedMemory {
return false
}
lastRun, ok := codexAutomationHeaderValue(lines[3], "Last run: ")
if !ok || !validCodexAutomationLastRun(lastRun) || lines[4] != "" {
return false
}
return strings.TrimSpace(strings.Join(lines[5:], "\n")) != ""
}
func codexAutomationHeaderValue(line, prefix string) (string, bool) {
if !strings.HasPrefix(line, prefix) {
return "", false
}
value := strings.TrimPrefix(line, prefix)
return value, value != "" && strings.TrimSpace(value) == value
}
func validCodexAutomationID(value string) bool {
if len(value) == 0 || len(value) > 128 || value == "." || value == ".." {
return false
}
for i := 0; i < len(value); i++ {
c := value[i]
if (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') || (c >= '0' && c <= '9') || c == '-' || c == '_' || c == '.' {
continue
}
return false
}
return true
}
func validCodexAutomationLastRun(value string) bool {
if value == "never" {
return true
}
separator := strings.LastIndex(value, " (")
if separator <= 0 || !strings.HasSuffix(value, ")") {
return false
}
runAt, err := time.Parse(time.RFC3339Nano, value[:separator])
if err != nil {
return false
}
epochMillis, err := strconv.ParseInt(value[separator+2:len(value)-1], 10, 64)
return err == nil && runAt.UnixMilli() == epochMillis
}
func validCodexDelegationEnvelope(value string) bool {
decoder := xml.NewDecoder(strings.NewReader(value))
var rootSeen, sourceSeen, inputSeen bool
var childName string
var childText bytes.Buffer
depth := 0
for {
token, err := decoder.Token()
if err == io.EOF {
return rootSeen && depth == 0 && sourceSeen && inputSeen
}
if err != nil {
return false
}
switch current := token.(type) {
case xml.StartElement:
depth++
if current.Name.Space != "" || len(current.Attr) != 0 || (depth == 1 && current.Name.Local != "codex_delegation") || depth > 2 {
return false
}
if depth == 1 {
if rootSeen {
return false
}
rootSeen = true
continue
}
if current.Name.Local != "source_thread_id" && current.Name.Local != "input" {
return false
}
childName = current.Name.Local
childText.Reset()
case xml.EndElement:
if current.Name.Space != "" {
return false
}
if depth == 2 {
if current.Name.Local != childName || strings.TrimSpace(childText.String()) == "" {
return false
}
if childName == "source_thread_id" {
if sourceSeen {
return false
}
sourceSeen = true
} else {
if inputSeen {
return false
}
inputSeen = true
}
childName = ""
}
depth--
if depth < 0 {
return false
}
case xml.CharData:
if depth == 2 {
_, _ = childText.Write(current)
} else if len(bytes.TrimSpace(current)) != 0 {
return false
}
case xml.Comment:
return false
case xml.ProcInst, xml.Directive:
return false
}
}
}
func (h *OpenAIGatewayHandler) acquireResponsesUserSlot(
c *gin.Context,
userID int64,
@@ -2298,7 +2622,7 @@ func (h *OpenAIGatewayHandler) ResponsesWebSocket(c *gin.Context) {
zap.Int("candidate_count", scheduleDecision.CandidateCount),
)
maxReasoningEffort, reasoningEffortMappings, _ := openAIReasoningEffortPolicyForRequest(c, apiKey)
maxReasoningEffort, reasoningEffortMappings, maxReasoningEffortOverLimit, _ := openAIReasoningEffortPolicyForRequest(c, apiKey)
var requestPayloadHash string
var turnStartsMu sync.Mutex
turnStarts := make(map[int]time.Time, 4)
@@ -2325,12 +2649,13 @@ func (h *OpenAIGatewayHandler) ResponsesWebSocket(c *gin.Context) {
// passthrough 没有 BeforeTurn 时，AfterTurn 回退到 TurnStarted 的所属 turn 时刻。
var turnPricing openAIWSTurnPricing
hooks := &service.OpenAIWSIngressHooks{
ClientLifecycleContext: clientLifecycleCtx,
InitialRequestModel: reqModel,
InitialTurnStartedAt: firstTurnStartedAt,
MaxReasoningEffort: maxReasoningEffort,
ReasoningEffortMappings: reasoningEffortMappings,
TurnStarted: recordTurnStart,
ClientLifecycleContext: clientLifecycleCtx,
InitialRequestModel: reqModel,
InitialTurnStartedAt: firstTurnStartedAt,
MaxReasoningEffort: maxReasoningEffort,
MaxReasoningEffortOverLimit: maxReasoningEffortOverLimit,
ReasoningEffortMappings: reasoningEffortMappings,
TurnStarted: recordTurnStart,
BeforeRequest: func(turn int, payload []byte, originalModel string) error {
c.Set(securityAuditWSTurnContextKey, turn)
service.BeginOpsStreamTurn(c, turn)
backend/internal/handler/ops_error_logger.go
+1
View File
@@ -2287,6 +2287,7 @@ func isOpsLocalBusinessLimitError(code string, msg string) bool {
strings.Contains(msg, "this group is restricted to claude code clients") ||
strings.Contains(msg, "this group does not allow /v1/messages dispatch") ||
strings.Contains(msg, "image generation is not enabled for this group") ||
(strings.Contains(msg, "reasoning effort") && strings.Contains(msg, "exceeds this group's limit")) ||
strings.Contains(msg, "token counting is not supported for this platform") ||
strings.Contains(msg, "images api is not supported for this platform") ||
(strings.Contains(msg, "model ") && strings.Contains(msg, " not in whitelist")) ||
backend/internal/handler/ops_error_logger_test.go
+15
View File
@@ -1243,6 +1243,15 @@ func TestClassifyOpsLocalBusinessLimitErrorsExcludedFromSLA(t *testing.T) {
wantErrType: "permission_error",
wantPhase: "request",
},
{
name: "group reasoning effort over limit deny",
errType: "permission_error",
message: `reasoning effort "high" exceeds this group's limit of "low"`,
code: "",
status: http.StatusForbidden,
wantErrType: "permission_error",
wantPhase: "request",
},
{
name: "route token counting platform unsupported",
errType: "not_found_error",
@@ -1504,6 +1513,12 @@ func TestClassifyOpsUpstreamAuthTextStillCountsForSLA(t *testing.T) {
code: "403",
status: http.StatusForbidden,
},
{
name: "provider reasoning effort over limit shaped error",
message: `reasoning effort "high" exceeds this group's limit of "low"`,
code: "403",
status: http.StatusForbidden,
},
{
name: "provider token counting unsupported shaped error",
message: "Token counting is not supported for this platform",