Merge branch 'Wei-Shaw:main' into fix/open-issues-cleanup

This commit is contained in:
IanShaw
2026-03-15 17:49:20 +08:00
committed by GitHub
38 changed files with 1219 additions and 68 deletions

View File

@@ -523,6 +523,8 @@ func usageLogFromServiceUser(l *service.UsageLog) UsageLog {
Model: l.Model,
ServiceTier: l.ServiceTier,
ReasoningEffort: l.ReasoningEffort,
InboundEndpoint: l.InboundEndpoint,
UpstreamEndpoint: l.UpstreamEndpoint,
GroupID: l.GroupID,
SubscriptionID: l.SubscriptionID,
InputTokens: l.InputTokens,

View File

@@ -76,10 +76,14 @@ func TestUsageLogFromService_IncludesServiceTierForUserAndAdmin(t *testing.T) {
t.Parallel()
serviceTier := "priority"
inboundEndpoint := "/v1/chat/completions"
upstreamEndpoint := "/v1/responses"
log := &service.UsageLog{
RequestID: "req_3",
Model: "gpt-5.4",
ServiceTier: &serviceTier,
InboundEndpoint: &inboundEndpoint,
UpstreamEndpoint: &upstreamEndpoint,
AccountRateMultiplier: f64Ptr(1.5),
}
@@ -88,8 +92,16 @@ func TestUsageLogFromService_IncludesServiceTierForUserAndAdmin(t *testing.T) {
require.NotNil(t, userDTO.ServiceTier)
require.Equal(t, serviceTier, *userDTO.ServiceTier)
require.NotNil(t, userDTO.InboundEndpoint)
require.Equal(t, inboundEndpoint, *userDTO.InboundEndpoint)
require.NotNil(t, userDTO.UpstreamEndpoint)
require.Equal(t, upstreamEndpoint, *userDTO.UpstreamEndpoint)
require.NotNil(t, adminDTO.ServiceTier)
require.Equal(t, serviceTier, *adminDTO.ServiceTier)
require.NotNil(t, adminDTO.InboundEndpoint)
require.Equal(t, inboundEndpoint, *adminDTO.InboundEndpoint)
require.NotNil(t, adminDTO.UpstreamEndpoint)
require.Equal(t, upstreamEndpoint, *adminDTO.UpstreamEndpoint)
require.NotNil(t, adminDTO.AccountRateMultiplier)
require.InDelta(t, 1.5, *adminDTO.AccountRateMultiplier, 1e-12)
}

View File

@@ -334,9 +334,13 @@ type UsageLog struct {
Model string `json:"model"`
// ServiceTier records the OpenAI service tier used for billing, e.g. "priority" / "flex".
ServiceTier *string `json:"service_tier,omitempty"`
// ReasoningEffort is the request's reasoning effort level (OpenAI Responses API).
// nil means not provided / not applicable.
// ReasoningEffort is the request's reasoning effort level.
// OpenAI: "low"/"medium"/"high"/"xhigh"; Claude: "low"/"medium"/"high"/"max".
ReasoningEffort *string `json:"reasoning_effort,omitempty"`
// InboundEndpoint is the client-facing API endpoint path, e.g. /v1/chat/completions.
InboundEndpoint *string `json:"inbound_endpoint,omitempty"`
// UpstreamEndpoint is the normalized upstream endpoint path, e.g. /v1/responses.
UpstreamEndpoint *string `json:"upstream_endpoint,omitempty"`
GroupID *int64 `json:"group_id"`
SubscriptionID *int64 `json:"subscription_id"`

View File

@@ -391,6 +391,8 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
if fs.SwitchCount > 0 {
requestCtx = service.WithAccountSwitchCount(requestCtx, fs.SwitchCount, h.metadataBridgeEnabled())
}
// 记录 Forward 前已写入字节数Forward 后若增加则说明 SSE 内容已发,禁止 failover
writerSizeBeforeForward := c.Writer.Size()
if account.Platform == service.PlatformAntigravity {
result, err = h.antigravityGatewayService.ForwardGemini(requestCtx, c, account, reqModel, "generateContent", reqStream, body, hasBoundSession)
} else {
@@ -402,6 +404,11 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
if err != nil {
var failoverErr *service.UpstreamFailoverError
if errors.As(err, &failoverErr) {
// 流式内容已写入客户端,无法撤销,禁止 failover 以防止流拼接腐化
if c.Writer.Size() != writerSizeBeforeForward {
h.handleFailoverExhausted(c, failoverErr, service.PlatformGemini, true)
return
}
action := fs.HandleFailoverError(c.Request.Context(), h.gatewayService, account.ID, account.Platform, failoverErr)
switch action {
case FailoverContinue:
@@ -436,6 +443,10 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
clientIP := ip.GetClientIP(c)
requestPayloadHash := service.HashUsageRequestPayload(body)
if result.ReasoningEffort == nil {
result.ReasoningEffort = service.NormalizeClaudeOutputEffort(parsedReq.OutputEffort)
}
// 使用量记录通过有界 worker 池提交,避免请求热路径创建无界 goroutine。
h.submitUsageRecordTask(func(ctx context.Context) {
if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{
@@ -637,6 +648,8 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
if fs.SwitchCount > 0 {
requestCtx = service.WithAccountSwitchCount(requestCtx, fs.SwitchCount, h.metadataBridgeEnabled())
}
// 记录 Forward 前已写入字节数Forward 后若增加则说明 SSE 内容已发,禁止 failover
writerSizeBeforeForward := c.Writer.Size()
if account.Platform == service.PlatformAntigravity && account.Type != service.AccountTypeAPIKey {
result, err = h.antigravityGatewayService.Forward(requestCtx, c, account, body, hasBoundSession)
} else {
@@ -706,6 +719,11 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
}
var failoverErr *service.UpstreamFailoverError
if errors.As(err, &failoverErr) {
// 流式内容已写入客户端,无法撤销,禁止 failover 以防止流拼接腐化
if c.Writer.Size() != writerSizeBeforeForward {
h.handleFailoverExhausted(c, failoverErr, account.Platform, true)
return
}
action := fs.HandleFailoverError(c.Request.Context(), h.gatewayService, account.ID, account.Platform, failoverErr)
switch action {
case FailoverContinue:
@@ -740,6 +758,10 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
clientIP := ip.GetClientIP(c)
requestPayloadHash := service.HashUsageRequestPayload(body)
if result.ReasoningEffort == nil {
result.ReasoningEffort = service.NormalizeClaudeOutputEffort(parsedReq.OutputEffort)
}
// 使用量记录通过有界 worker 池提交,避免请求热路径创建无界 goroutine。
h.submitUsageRecordTask(func(ctx context.Context) {
if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{

View File

@@ -0,0 +1,122 @@
package handler
import (
"net/http"
"net/http/httptest"
"strings"
"testing"
"github.com/Wei-Shaw/sub2api/internal/service"
"github.com/gin-gonic/gin"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// partialMessageStartSSE 模拟 handleStreamingResponse 已写入的首批 SSE 事件。
const partialMessageStartSSE = "event: message_start\ndata: {\"type\":\"message_start\",\"message\":{\"id\":\"msg_01\",\"type\":\"message\",\"role\":\"assistant\",\"content\":[],\"model\":\"claude-sonnet-4-5\",\"stop_reason\":null,\"stop_sequence\":null,\"usage\":{\"input_tokens\":10,\"output_tokens\":1}}}\n\n" +
"event: content_block_start\ndata: {\"type\":\"content_block_start\",\"index\":0,\"content_block\":{\"type\":\"text\",\"text\":\"\"}}\n\n"
// TestStreamWrittenGuard_MessagesPath_AbortFailoverOnSSEContentWritten 验证:
// 当 Forward 在返回 UpstreamFailoverError 前已向客户端写入 SSE 内容时,
// 故障转移保护逻辑必须终止循环并发送 SSE 错误事件,而不是进行下一次 Forward。
// 具体验证:
// 1. c.Writer.Size() 检测条件正确触发(字节数已增加)
// 2. handleFailoverExhausted 以 streamStarted=true 调用后,响应体以 SSE 错误事件结尾
// 3. 响应体中只出现一个 message_start不存在第二个防止流拼接腐化
func TestStreamWrittenGuard_MessagesPath_AbortFailoverOnSSEContentWritten(t *testing.T) {
gin.SetMode(gin.TestMode)
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", nil)
// 步骤 1记录 Forward 前的 writer size模拟 writerSizeBeforeForward := c.Writer.Size()
sizeBeforeForward := c.Writer.Size()
require.Equal(t, -1, sizeBeforeForward, "gin writer 初始 Size 应为 -1未写入任何字节")
// 步骤 2模拟 Forward 已向客户端写入部分 SSE 内容message_start + content_block_start
_, err := c.Writer.Write([]byte(partialMessageStartSSE))
require.NoError(t, err)
// 步骤 3验证守卫条件成立c.Writer.Size() != sizeBeforeForward
require.NotEqual(t, sizeBeforeForward, c.Writer.Size(),
"写入 SSE 内容后 writer size 必须增加,守卫条件应为 true")
// 步骤 4模拟 UpstreamFailoverError上游在流中途返回 403
failoverErr := &service.UpstreamFailoverError{
StatusCode: http.StatusForbidden,
ResponseBody: []byte(`{"error":{"type":"permission_error","message":"forbidden"}}`),
}
// 步骤 5守卫触发 → 调用 handleFailoverExhaustedstreamStarted=true
h := &GatewayHandler{}
h.handleFailoverExhausted(c, failoverErr, service.PlatformAnthropic, true)
body := w.Body.String()
// 断言 A响应体中包含最初写入的 message_start SSE 事件行
require.Contains(t, body, "event: message_start", "响应体应包含已写入的 message_start SSE 事件")
// 断言 B响应体以 SSE 错误事件结尾data: {"type":"error",...}\n\n
require.True(t, strings.HasSuffix(strings.TrimRight(body, "\n"), "}"),
"响应体应以 JSON 对象结尾SSE error event 的 data 字段)")
require.Contains(t, body, `"type":"error"`, "响应体末尾必须包含 SSE 错误事件")
// 断言 CSSE event 行 "event: message_start" 只出现一次(防止双 message_start 拼接腐化)
firstIdx := strings.Index(body, "event: message_start")
lastIdx := strings.LastIndex(body, "event: message_start")
assert.Equal(t, firstIdx, lastIdx,
"响应体中 'event: message_start' 必须只出现一次,不得因 failover 拼接导致两次")
}
// TestStreamWrittenGuard_GeminiPath_AbortFailoverOnSSEContentWritten 与上述测试相同,
// 验证 Gemini 路径使用 service.PlatformGemini而非 account.Platform时行为一致。
func TestStreamWrittenGuard_GeminiPath_AbortFailoverOnSSEContentWritten(t *testing.T) {
gin.SetMode(gin.TestMode)
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
c.Request = httptest.NewRequest(http.MethodPost, "/v1beta/models/gemini-2.0-flash:streamGenerateContent", nil)
sizeBeforeForward := c.Writer.Size()
_, err := c.Writer.Write([]byte(partialMessageStartSSE))
require.NoError(t, err)
require.NotEqual(t, sizeBeforeForward, c.Writer.Size())
failoverErr := &service.UpstreamFailoverError{
StatusCode: http.StatusForbidden,
}
h := &GatewayHandler{}
h.handleFailoverExhausted(c, failoverErr, service.PlatformGemini, true)
body := w.Body.String()
require.Contains(t, body, "event: message_start")
require.Contains(t, body, `"type":"error"`)
firstIdx := strings.Index(body, "event: message_start")
lastIdx := strings.LastIndex(body, "event: message_start")
assert.Equal(t, firstIdx, lastIdx, "Gemini 路径不得出现双 message_start")
}
// TestStreamWrittenGuard_NoByteWritten_GuardNotTriggered 验证反向场景:
// 当 Forward 返回 UpstreamFailoverError 时若未向客户端写入任何 SSE 内容,
// 守卫条件c.Writer.Size() != sizeBeforeForward为 false不应中止 failover。
func TestStreamWrittenGuard_NoByteWritten_GuardNotTriggered(t *testing.T) {
gin.SetMode(gin.TestMode)
w := httptest.NewRecorder()
c, _ := gin.CreateTestContext(w)
c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", nil)
// 模拟 writerSizeBeforeForward初始为 -1
sizeBeforeForward := c.Writer.Size()
// Forward 未写入任何字节直接返回错误(例如 401 发生在连接建立前)
// c.Writer.Size() 仍为 -1
// 守卫条件sizeBeforeForward == c.Writer.Size() → 不触发
guardTriggered := c.Writer.Size() != sizeBeforeForward
require.False(t, guardTriggered,
"未写入任何字节时,守卫条件必须为 false应允许正常 failover 继续")
}

View File

@@ -256,14 +256,16 @@ func (h *OpenAIGatewayHandler) ChatCompletions(c *gin.Context) {
h.submitUsageRecordTask(func(ctx context.Context) {
if err := h.gatewayService.RecordUsage(ctx, &service.OpenAIRecordUsageInput{
Result: result,
APIKey: apiKey,
User: apiKey.User,
Account: account,
Subscription: subscription,
UserAgent: userAgent,
IPAddress: clientIP,
APIKeyService: h.apiKeyService,
Result: result,
APIKey: apiKey,
User: apiKey.User,
Account: account,
Subscription: subscription,
InboundEndpoint: normalizedOpenAIInboundEndpoint(c, openAIInboundEndpointChatCompletions),
UpstreamEndpoint: normalizedOpenAIUpstreamEndpoint(c, openAIUpstreamEndpointResponses),
UserAgent: userAgent,
IPAddress: clientIP,
APIKeyService: h.apiKeyService,
}); err != nil {
logger.L().With(
zap.String("component", "handler.openai_gateway.chat_completions"),

View File

@@ -0,0 +1,57 @@
package handler
import (
"net/http"
"net/http/httptest"
"testing"
"github.com/gin-gonic/gin"
"github.com/stretchr/testify/require"
)
func TestNormalizedOpenAIUpstreamEndpoint(t *testing.T) {
gin.SetMode(gin.TestMode)
tests := []struct {
name string
path string
fallback string
want string
}{
{
name: "responses root maps to responses upstream",
path: "/v1/responses",
fallback: openAIUpstreamEndpointResponses,
want: "/v1/responses",
},
{
name: "responses compact keeps compact suffix",
path: "/openai/v1/responses/compact",
fallback: openAIUpstreamEndpointResponses,
want: "/v1/responses/compact",
},
{
name: "responses nested suffix preserved",
path: "/openai/v1/responses/compact/detail",
fallback: openAIUpstreamEndpointResponses,
want: "/v1/responses/compact/detail",
},
{
name: "non responses path uses fallback",
path: "/v1/messages",
fallback: openAIUpstreamEndpointResponses,
want: "/v1/responses",
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
rec := httptest.NewRecorder()
c, _ := gin.CreateTestContext(rec)
c.Request = httptest.NewRequest(http.MethodPost, tt.path, nil)
got := normalizedOpenAIUpstreamEndpoint(c, tt.fallback)
require.Equal(t, tt.want, got)
})
}
}

View File

@@ -37,6 +37,13 @@ type OpenAIGatewayHandler struct {
cfg *config.Config
}
const (
openAIInboundEndpointResponses = "/v1/responses"
openAIInboundEndpointMessages = "/v1/messages"
openAIInboundEndpointChatCompletions = "/v1/chat/completions"
openAIUpstreamEndpointResponses = "/v1/responses"
)
// NewOpenAIGatewayHandler creates a new OpenAIGatewayHandler
func NewOpenAIGatewayHandler(
gatewayService *service.OpenAIGatewayService,
@@ -362,6 +369,8 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
User: apiKey.User,
Account: account,
Subscription: subscription,
InboundEndpoint: normalizedOpenAIInboundEndpoint(c, openAIInboundEndpointResponses),
UpstreamEndpoint: normalizedOpenAIUpstreamEndpoint(c, openAIUpstreamEndpointResponses),
UserAgent: userAgent,
IPAddress: clientIP,
RequestPayloadHash: requestPayloadHash,
@@ -738,6 +747,8 @@ func (h *OpenAIGatewayHandler) Messages(c *gin.Context) {
User: apiKey.User,
Account: account,
Subscription: subscription,
InboundEndpoint: normalizedOpenAIInboundEndpoint(c, openAIInboundEndpointMessages),
UpstreamEndpoint: normalizedOpenAIUpstreamEndpoint(c, openAIUpstreamEndpointResponses),
UserAgent: userAgent,
IPAddress: clientIP,
RequestPayloadHash: requestPayloadHash,
@@ -1235,6 +1246,8 @@ func (h *OpenAIGatewayHandler) ResponsesWebSocket(c *gin.Context) {
User: apiKey.User,
Account: account,
Subscription: subscription,
InboundEndpoint: normalizedOpenAIInboundEndpoint(c, openAIInboundEndpointResponses),
UpstreamEndpoint: normalizedOpenAIUpstreamEndpoint(c, openAIUpstreamEndpointResponses),
UserAgent: userAgent,
IPAddress: clientIP,
RequestPayloadHash: service.HashUsageRequestPayload(firstMessage),
@@ -1530,6 +1543,62 @@ func openAIWSIngressFallbackSessionSeed(userID, apiKeyID int64, groupID *int64)
return fmt.Sprintf("openai_ws_ingress:%d:%d:%d", gid, userID, apiKeyID)
}
func normalizedOpenAIInboundEndpoint(c *gin.Context, fallback string) string {
path := strings.TrimSpace(fallback)
if c != nil {
if fullPath := strings.TrimSpace(c.FullPath()); fullPath != "" {
path = fullPath
} else if c.Request != nil && c.Request.URL != nil {
if requestPath := strings.TrimSpace(c.Request.URL.Path); requestPath != "" {
path = requestPath
}
}
}
switch {
case strings.Contains(path, openAIInboundEndpointChatCompletions):
return openAIInboundEndpointChatCompletions
case strings.Contains(path, openAIInboundEndpointMessages):
return openAIInboundEndpointMessages
case strings.Contains(path, openAIInboundEndpointResponses):
return openAIInboundEndpointResponses
default:
return path
}
}
func normalizedOpenAIUpstreamEndpoint(c *gin.Context, fallback string) string {
base := strings.TrimSpace(fallback)
if base == "" {
base = openAIUpstreamEndpointResponses
}
base = strings.TrimRight(base, "/")
if c == nil || c.Request == nil || c.Request.URL == nil {
return base
}
path := strings.TrimRight(strings.TrimSpace(c.Request.URL.Path), "/")
if path == "" {
return base
}
idx := strings.LastIndex(path, "/responses")
if idx < 0 {
return base
}
suffix := strings.TrimSpace(path[idx+len("/responses"):])
if suffix == "" || suffix == "/" {
return base
}
if !strings.HasPrefix(suffix, "/") {
return base
}
return base + suffix
}
func isOpenAIWSUpgradeRequest(r *http.Request) bool {
if r == nil {
return false

View File

@@ -334,6 +334,14 @@ func (s *stubUsageLogRepo) GetUsageTrendWithFilters(ctx context.Context, startTi
func (s *stubUsageLogRepo) GetModelStatsWithFilters(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, requestType *int16, stream *bool, billingType *int8) ([]usagestats.ModelStat, error) {
return nil, nil
}
func (s *stubUsageLogRepo) GetEndpointStatsWithFilters(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, model string, requestType *int16, stream *bool, billingType *int8) ([]usagestats.EndpointStat, error) {
return []usagestats.EndpointStat{}, nil
}
func (s *stubUsageLogRepo) GetUpstreamEndpointStatsWithFilters(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, model string, requestType *int16, stream *bool, billingType *int8) ([]usagestats.EndpointStat, error) {
return []usagestats.EndpointStat{}, nil
}
func (s *stubUsageLogRepo) GetGroupStatsWithFilters(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, requestType *int16, stream *bool, billingType *int8) ([]usagestats.GroupStat, error) {
return nil, nil
}