fix: drain OpenAI compat streams for usage

2026-05-03 17:11:27 +08:00
parent b2bdba78dd
commit 72d5ee4cd1
13 changed files with 1141 additions and 205 deletions
@@ -1,13 +1,36 @@
 package service

 import (
+	"bytes"
+	"context"
+	"errors"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
 	"testing"
+	"time"

 	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
+	"github.com/gin-gonic/gin"
 	"github.com/stretchr/testify/require"
 	"github.com/tidwall/gjson"
 )

+type openAIChatFailingWriter struct {
+	gin.ResponseWriter
+	failAfter int
+	writes    int
+}
+
+func (w *openAIChatFailingWriter) Write(p []byte) (int, error) {
+	if w.writes >= w.failAfter {
+		return 0, errors.New("write failed: client disconnected")
+	}
+	w.writes++
+	return w.ResponseWriter.Write(p)
+}
+
 func TestNormalizeResponsesRequestServiceTier(t *testing.T) {
 	t.Parallel()

@@ -73,3 +96,238 @@ func TestNormalizeResponsesBodyServiceTier(t *testing.T) {
 	require.Empty(t, tier)
 	require.False(t, gjson.GetBytes(body, "service_tier").Exists())
 }
+
+func TestForwardAsChatCompletions_ClientDisconnectDrainsUpstreamUsage(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Writer = &openAIChatFailingWriter{ResponseWriter: c.Writer, failAfter: 0}
+	body := []byte(`{"model":"gpt-5.4","messages":[{"role":"user","content":"hello"}],"stream":true}`)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/chat/completions", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	upstreamBody := strings.Join([]string{
+		`data: {"type":"response.created","response":{"id":"resp_1","model":"gpt-5.4","status":"in_progress","output":[]}}`,
+		"",
+		`data: {"type":"response.output_text.delta","delta":"ok"}`,
+		"",
+		`data: {"type":"response.completed","response":{"id":"resp_1","object":"response","model":"gpt-5.4","status":"completed","output":[{"type":"message","id":"msg_1","role":"assistant","status":"completed","content":[{"type":"output_text","text":"ok"}]}],"usage":{"input_tokens":11,"output_tokens":5,"total_tokens":16,"input_tokens_details":{"cached_tokens":4}}}}`,
+		"",
+		"data: [DONE]",
+		"",
+	}, "\n")
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}, "x-request-id": []string{"rid_chat_disconnect"}},
+		Body:       io.NopCloser(strings.NewReader(upstreamBody)),
+	}}
+
+	svc := &OpenAIGatewayService{httpUpstream: upstream}
+	account := &Account{
+		ID:          1,
+		Name:        "openai-oauth",
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeOAuth,
+		Concurrency: 1,
+		Credentials: map[string]any{
+			"access_token":       "oauth-token",
+			"chatgpt_account_id": "chatgpt-acc",
+		},
+	}
+
+	result, err := svc.ForwardAsChatCompletions(context.Background(), c, account, body, "", "gpt-5.1")
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.Equal(t, 11, result.Usage.InputTokens)
+	require.Equal(t, 5, result.Usage.OutputTokens)
+	require.Equal(t, 4, result.Usage.CacheReadInputTokens)
+}
+
+func TestForwardAsChatCompletions_TerminalUsageWithoutUpstreamCloseReturns(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Writer = &openAIChatFailingWriter{ResponseWriter: c.Writer, failAfter: 0}
+	body := []byte(`{"model":"gpt-5.4","messages":[{"role":"user","content":"hello"}],"stream":true}`)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/chat/completions", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	upstreamBody := []byte(`data: {"type":"response.completed","response":{"id":"resp_1","object":"response","model":"gpt-5.4","status":"completed","output":[{"type":"message","id":"msg_1","role":"assistant","status":"completed","content":[{"type":"output_text","text":"ok"}]}],"usage":{"input_tokens":17,"output_tokens":8,"total_tokens":25,"input_tokens_details":{"cached_tokens":6}}}}` + "\n\n")
+	upstreamStream := newOpenAICompatBlockingReadCloser(upstreamBody)
+	defer upstreamStream.Close()
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}, "x-request-id": []string{"rid_chat_terminal_no_close"}},
+		Body:       upstreamStream,
+	}}
+
+	svc := &OpenAIGatewayService{httpUpstream: upstream}
+	account := &Account{
+		ID:          1,
+		Name:        "openai-oauth",
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeOAuth,
+		Concurrency: 1,
+		Credentials: map[string]any{
+			"access_token":       "oauth-token",
+			"chatgpt_account_id": "chatgpt-acc",
+		},
+	}
+
+	type forwardResult struct {
+		result *OpenAIForwardResult
+		err    error
+	}
+	resultCh := make(chan forwardResult, 1)
+	go func() {
+		result, err := svc.ForwardAsChatCompletions(context.Background(), c, account, body, "", "gpt-5.1")
+		resultCh <- forwardResult{result: result, err: err}
+	}()
+
+	select {
+	case got := <-resultCh:
+		require.NoError(t, got.err)
+		require.NotNil(t, got.result)
+		require.Equal(t, 17, got.result.Usage.InputTokens)
+		require.Equal(t, 8, got.result.Usage.OutputTokens)
+		require.Equal(t, 6, got.result.Usage.CacheReadInputTokens)
+	case <-time.After(time.Second):
+		require.Fail(t, "ForwardAsChatCompletions should return after terminal usage event even if upstream keeps the connection open")
+	}
+}
+
+func TestForwardAsChatCompletions_BufferedTerminalWithoutUpstreamCloseReturns(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	body := []byte(`{"model":"gpt-5.4","messages":[{"role":"user","content":"hello"}],"stream":false}`)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/chat/completions", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	upstreamBody := []byte(`data: {"type":"response.completed","response":{"id":"resp_1","object":"response","model":"gpt-5.4","status":"completed","output":[{"type":"message","id":"msg_1","role":"assistant","status":"completed","content":[{"type":"output_text","text":"ok"}]}],"usage":{"input_tokens":17,"output_tokens":8,"total_tokens":25,"input_tokens_details":{"cached_tokens":6}}}}` + "\n\n")
+	upstreamStream := newOpenAICompatBlockingReadCloser(upstreamBody)
+	defer upstreamStream.Close()
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}, "x-request-id": []string{"rid_chat_buffered_terminal_no_close"}},
+		Body:       upstreamStream,
+	}}
+
+	svc := &OpenAIGatewayService{httpUpstream: upstream}
+	account := &Account{
+		ID:          1,
+		Name:        "openai-oauth",
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeOAuth,
+		Concurrency: 1,
+		Credentials: map[string]any{
+			"access_token":       "oauth-token",
+			"chatgpt_account_id": "chatgpt-acc",
+		},
+	}
+
+	type forwardResult struct {
+		result *OpenAIForwardResult
+		err    error
+	}
+	resultCh := make(chan forwardResult, 1)
+	go func() {
+		result, err := svc.ForwardAsChatCompletions(context.Background(), c, account, body, "", "gpt-5.1")
+		resultCh <- forwardResult{result: result, err: err}
+	}()
+
+	select {
+	case got := <-resultCh:
+		require.NoError(t, got.err)
+		require.NotNil(t, got.result)
+		require.Equal(t, 17, got.result.Usage.InputTokens)
+		require.Equal(t, 8, got.result.Usage.OutputTokens)
+		require.Equal(t, 6, got.result.Usage.CacheReadInputTokens)
+		require.Contains(t, rec.Body.String(), `"finish_reason":"stop"`)
+	case <-time.After(time.Second):
+		require.Fail(t, "ForwardAsChatCompletions buffered response should return after terminal usage event even if upstream keeps the connection open")
+	}
+}
+
+func TestForwardAsChatCompletions_DoneSentinelWithoutTerminalReturnsError(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	body := []byte(`{"model":"gpt-5.4","messages":[{"role":"user","content":"hello"}],"stream":true}`)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/chat/completions", bytes.NewReader(body))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	upstreamBody := "data: [DONE]\n\n"
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}, "x-request-id": []string{"rid_chat_missing_terminal"}},
+		Body:       io.NopCloser(strings.NewReader(upstreamBody)),
+	}}
+
+	svc := &OpenAIGatewayService{httpUpstream: upstream}
+	account := &Account{
+		ID:          1,
+		Name:        "openai-oauth",
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeOAuth,
+		Concurrency: 1,
+		Credentials: map[string]any{
+			"access_token":       "oauth-token",
+			"chatgpt_account_id": "chatgpt-acc",
+		},
+	}
+
+	result, err := svc.ForwardAsChatCompletions(context.Background(), c, account, body, "", "gpt-5.1")
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "missing terminal event")
+	require.NotNil(t, result)
+	require.Zero(t, result.Usage.InputTokens)
+	require.Zero(t, result.Usage.OutputTokens)
+}
+
+func TestForwardAsChatCompletions_UpstreamRequestIgnoresClientCancel(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	reqCtx, cancel := context.WithCancel(context.Background())
+	body := []byte(`{"model":"gpt-5.4","messages":[{"role":"user","content":"hello"}],"stream":false}`)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/chat/completions", bytes.NewReader(body)).WithContext(reqCtx)
+	c.Request.Header.Set("Content-Type", "application/json")
+	cancel()
+
+	upstreamBody := strings.Join([]string{
+		`data: {"type":"response.completed","response":{"id":"resp_1","object":"response","model":"gpt-5.4","status":"completed","output":[{"type":"message","id":"msg_1","role":"assistant","status":"completed","content":[{"type":"output_text","text":"ok"}]}],"usage":{"input_tokens":5,"output_tokens":2,"total_tokens":7}}}`,
+		"",
+		"data: [DONE]",
+		"",
+	}, "\n")
+	upstream := &httpUpstreamRecorder{resp: &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}, "x-request-id": []string{"rid_chat_ctx"}},
+		Body:       io.NopCloser(strings.NewReader(upstreamBody)),
+	}}
+
+	svc := &OpenAIGatewayService{httpUpstream: upstream}
+	account := &Account{
+		ID:          1,
+		Name:        "openai-oauth",
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeOAuth,
+		Concurrency: 1,
+		Credentials: map[string]any{
+			"access_token":       "oauth-token",
+			"chatgpt_account_id": "chatgpt-acc",
+		},
+	}
+
+	result, err := svc.ForwardAsChatCompletions(reqCtx, c, account, body, "", "gpt-5.1")
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.NotNil(t, upstream.lastReq)
+	require.NoError(t, upstream.lastReq.Context().Err())
+}