Merge pull request #4296 from wp-a/fix/openai-responses-sse-json-boundaries

fix(openai): repair concatenated Responses stream events
This commit is contained in:
Wesley Liddick
2026-07-15 10:11:55 +08:00
committed by GitHub
5 changed files with 423 additions and 11 deletions
@@ -971,6 +971,7 @@ func (s *OpenAIGatewayService) handleStreamingResponsePassthrough(
scanBuf := getSSEScannerBuf64K()
scanner.Buffer(scanBuf[:0], maxLineSize)
defer putSSEScannerBuf64K(scanBuf)
documentScanner := newOpenAISSEJSONDocumentScanner(scanner)
needModelReplace := strings.TrimSpace(originalModel) != "" && strings.TrimSpace(mappedModel) != "" && strings.TrimSpace(originalModel) != strings.TrimSpace(mappedModel)
resultWithUsage := func() *openaiStreamingResultPassthrough {
@@ -983,8 +984,8 @@ func (s *OpenAIGatewayService) handleStreamingResponsePassthrough(
}
}
for scanner.Scan() {
line := scanner.Text()
for documentScanner.Scan() {
line := documentScanner.Text()
lineStartsClientOutput := false
forceFlushFailedEvent := false
if data, ok := extractOpenAISSEDataLine(line); ok {
@@ -1104,7 +1105,7 @@ func (s *OpenAIGatewayService) handleStreamingResponsePassthrough(
}
}
}
if err := scanner.Err(); err != nil {
if err := documentScanner.Err(); err != nil {
if sawTerminalEvent && !sawFailedEvent {
return resultWithUsage(), nil
}
@@ -79,6 +79,7 @@ func (s *OpenAIGatewayService) handleStreamingResponse(ctx context.Context, resp
}
scanBuf := getSSEScannerBuf64K()
scanner.Buffer(scanBuf[:0], maxLineSize)
documentScanner := newOpenAISSEJSONDocumentScanner(scanner)
streamInterval := time.Duration(0)
if s.cfg != nil && s.cfg.Gateway.StreamDataIntervalTimeout > 0 {
@@ -393,13 +394,13 @@ func (s *OpenAIGatewayService) handleStreamingResponse(ctx context.Context, resp
// 无超时/无 keepalive 的常见路径走同步扫描,减少 goroutine 与 channel 开销。
if streamInterval <= 0 && keepaliveInterval <= 0 {
defer putSSEScannerBuf64K(scanBuf)
for scanner.Scan() {
processSSELine(scanner.Text(), true)
for documentScanner.Scan() {
processSSELine(documentScanner.Text(), true)
if streamEarlyErr != nil {
return resultWithUsage(), streamEarlyErr
}
}
if result, err, done := handleScanErr(scanner.Err()); done {
if result, err, done := handleScanErr(documentScanner.Err()); done {
return result, err
}
return finalizeStream()
@@ -425,13 +426,13 @@ func (s *OpenAIGatewayService) handleStreamingResponse(ctx context.Context, resp
go func(scanBuf *sseScannerBuf64K) {
defer putSSEScannerBuf64K(scanBuf)
defer close(events)
for scanner.Scan() {
for documentScanner.Scan() {
atomic.StoreInt64(&lastReadAt, time.Now().UnixNano())
if !sendEvent(scanEvent{line: scanner.Text()}) {
if !sendEvent(scanEvent{line: documentScanner.Text()}) {
return
}
}
if err := scanner.Err(); err != nil {
if err := documentScanner.Err(); err != nil {
_ = sendEvent(scanEvent{err: err})
}
}(scanBuf)
@@ -0,0 +1,271 @@
package service
import (
"bufio"
"context"
"encoding/json"
"io"
"net/http"
"net/http/httptest"
"strings"
"testing"
"time"
"github.com/Wei-Shaw/sub2api/internal/config"
"github.com/gin-gonic/gin"
"github.com/stretchr/testify/require"
)
func TestOpenAIStreamingRepairsConcatenatedJSONDocumentsInSingleDataLine(t *testing.T) {
testOpenAIStreamingRepairsConcatenatedJSONDocuments(t, false, 0)
}
func TestOpenAIStreamingAsyncScannerRepairsConcatenatedJSONDocumentsInSingleDataLine(t *testing.T) {
testOpenAIStreamingRepairsConcatenatedJSONDocuments(t, false, 30)
}
func TestOpenAIStreamingPassthroughRepairsConcatenatedJSONDocumentsInSingleDataLine(t *testing.T) {
testOpenAIStreamingRepairsConcatenatedJSONDocuments(t, true, 0)
}
func TestOpenAIWSv2StreamingRepairsConcatenatedJSONDocumentsInSingleMessage(t *testing.T) {
gin.SetMode(gin.TestMode)
largeInProgress, outputItemAdded, completed := openAIConcatenatedJSONTestEvents(t)
captureConn := &openAIWSCaptureConn{events: [][]byte{
[]byte(largeInProgress + outputItemAdded),
[]byte(completed),
}}
cfg := &config.Config{}
cfg.Security.URLAllowlist.Enabled = false
cfg.Gateway.OpenAIWS.Enabled = true
cfg.Gateway.OpenAIWS.APIKeyEnabled = true
cfg.Gateway.OpenAIWS.ResponsesWebsocketsV2 = true
cfg.Gateway.OpenAIWS.MaxConnsPerAccount = 1
cfg.Gateway.OpenAIWS.MaxIdlePerAccount = 1
cfg.Gateway.OpenAIWS.QueueLimitPerConn = 8
cfg.Gateway.OpenAIWS.DialTimeoutSeconds = 3
cfg.Gateway.OpenAIWS.ReadTimeoutSeconds = 5
cfg.Gateway.OpenAIWS.WriteTimeoutSeconds = 3
pool := newOpenAIWSConnPool(cfg)
pool.setClientDialerForTest(&openAIWSCaptureDialer{conn: captureConn})
svc := &OpenAIGatewayService{
cfg: cfg,
cache: &stubGatewayCache{},
httpUpstream: &httpUpstreamRecorder{},
openaiWSResolver: NewOpenAIWSProtocolResolver(cfg),
openaiWSPool: pool,
toolCorrector: NewCodexToolCorrector(),
}
account := &Account{
ID: 2,
Name: "ws-test",
Platform: PlatformOpenAI,
Type: AccountTypeAPIKey,
Status: StatusActive,
Schedulable: true,
Concurrency: 1,
Credentials: map[string]any{"api_key": "sk-test"},
Extra: map[string]any{"responses_websockets_v2_enabled": true},
}
recorder := httptest.NewRecorder()
c, _ := gin.CreateTestContext(recorder)
c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses", nil)
groupID := int64(1)
c.Set("api_key", &APIKey{GroupID: &groupID})
result, err := svc.Forward(context.Background(), c, account, []byte(`{"model":"gpt-5.6-sol","stream":true,"input":"hello"}`))
require.NoError(t, err)
require.NotNil(t, result)
require.Equal(t, 7, result.Usage.InputTokens)
require.Equal(t, 9, result.Usage.OutputTokens)
require.Nil(t, result.FirstTokenMs)
assertOpenAISSEFrames(t, recorder.Body.String(), []string{
"response.in_progress",
"response.output_item.added",
"response.completed",
})
}
func TestSplitOpenAIConcatenatedJSONDocumentsRejectsPayloadOverRepairLimit(t *testing.T) {
first := `{"type":"response.in_progress","padding":"` + strings.Repeat("x", 16*1024*1024) + `"}`
second := `{"type":"response.completed"}`
payload := first + second
documents, repaired := splitOpenAIConcatenatedJSONDocuments([]byte(payload))
require.False(t, repaired)
require.Nil(t, documents)
line := "data: " + payload
scanner := bufio.NewScanner(strings.NewReader(line))
scanner.Buffer(make([]byte, 1024), len(line)+1)
documentScanner := newOpenAISSEJSONDocumentScanner(scanner)
require.True(t, documentScanner.Scan())
require.Equal(t, line, documentScanner.Text())
require.False(t, documentScanner.Scan())
require.NoError(t, documentScanner.Err())
}
func TestOpenAIWSv2StreamingBreaksConnectionWhenTerminalHasTrailingDocument(t *testing.T) {
gin.SetMode(gin.TestMode)
completed := `{"type":"response.completed","response":{"id":"resp_terminal_tail","usage":{"input_tokens":2,"output_tokens":1}}}`
tail := `{"type":"error","error":{"type":"upstream_error","message":"tail"}}`
captureConn := &openAIWSCaptureConn{events: [][]byte{[]byte(completed + tail)}}
cfg := &config.Config{}
cfg.Security.URLAllowlist.Enabled = false
cfg.Gateway.OpenAIWS.Enabled = true
cfg.Gateway.OpenAIWS.APIKeyEnabled = true
cfg.Gateway.OpenAIWS.ResponsesWebsocketsV2 = true
cfg.Gateway.OpenAIWS.MaxConnsPerAccount = 1
cfg.Gateway.OpenAIWS.MaxIdlePerAccount = 1
cfg.Gateway.OpenAIWS.QueueLimitPerConn = 8
cfg.Gateway.OpenAIWS.DialTimeoutSeconds = 3
cfg.Gateway.OpenAIWS.ReadTimeoutSeconds = 5
cfg.Gateway.OpenAIWS.WriteTimeoutSeconds = 3
pool := newOpenAIWSConnPool(cfg)
pool.setClientDialerForTest(&openAIWSCaptureDialer{conn: captureConn})
svc := &OpenAIGatewayService{
cfg: cfg,
cache: &stubGatewayCache{},
httpUpstream: &httpUpstreamRecorder{},
openaiWSResolver: NewOpenAIWSProtocolResolver(cfg),
openaiWSPool: pool,
toolCorrector: NewCodexToolCorrector(),
}
account := &Account{
ID: 3,
Name: "ws-terminal-tail",
Platform: PlatformOpenAI,
Type: AccountTypeAPIKey,
Status: StatusActive,
Schedulable: true,
Concurrency: 1,
Credentials: map[string]any{"api_key": "sk-test"},
Extra: map[string]any{"responses_websockets_v2_enabled": true},
}
recorder := httptest.NewRecorder()
c, _ := gin.CreateTestContext(recorder)
c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses", nil)
groupID := int64(1)
c.Set("api_key", &APIKey{GroupID: &groupID})
result, err := svc.Forward(context.Background(), c, account, []byte(`{"model":"gpt-5.6-sol","stream":true,"input":"hello"}`))
require.NoError(t, err)
require.NotNil(t, result)
require.True(t, captureConn.closed, "a WS message with data after a terminal event must not return to the pool")
assertOpenAISSEFrames(t, recorder.Body.String(), []string{"response.completed"})
}
func testOpenAIStreamingRepairsConcatenatedJSONDocuments(t *testing.T, passthrough bool, streamDataIntervalTimeout int) {
t.Helper()
gin.SetMode(gin.TestMode)
largeInProgress, outputItemAdded, completed := openAIConcatenatedJSONTestEvents(t)
upstreamBody := strings.Join([]string{
"event: response.in_progress",
"data: " + largeInProgress + outputItemAdded,
"",
"event: response.completed",
"data: " + completed,
"",
}, "\n")
resp := &http.Response{
StatusCode: http.StatusOK,
Header: http.Header{"Content-Type": []string{"text/event-stream"}},
Body: io.NopCloser(strings.NewReader(upstreamBody)),
}
recorder := httptest.NewRecorder()
c, _ := gin.CreateTestContext(recorder)
c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses", nil)
svc := &OpenAIGatewayService{
cfg: &config.Config{Gateway: config.GatewayConfig{
MaxLineSize: defaultMaxLineSize,
StreamDataIntervalTimeout: streamDataIntervalTimeout,
}},
toolCorrector: NewCodexToolCorrector(),
}
account := &Account{ID: 1, Name: "test", Platform: PlatformOpenAI}
var usage *OpenAIUsage
var err error
if passthrough {
result, forwardErr := svc.handleStreamingResponsePassthrough(c.Request.Context(), resp, c, account, time.Now(), "gpt-5.6-sol", "gpt-5.6-sol")
err = forwardErr
if result != nil {
usage = result.usage
}
} else {
result, forwardErr := svc.handleStreamingResponse(c.Request.Context(), resp, c, account, time.Now(), "gpt-5.6-sol", "gpt-5.6-sol")
err = forwardErr
if result != nil {
usage = result.usage
}
}
require.NoError(t, err)
require.NotNil(t, usage)
require.Equal(t, 7, usage.InputTokens)
require.Equal(t, 9, usage.OutputTokens)
assertOpenAISSEFrames(t, recorder.Body.String(), []string{
"response.in_progress",
"response.output_item.added",
"response.completed",
})
}
func assertOpenAISSEFrames(t *testing.T, body string, expectedTypes []string) {
t.Helper()
var parser openAICompatSSEFrameParser
var eventTypes []string
for _, line := range strings.Split(body, "\n") {
frame, ok := parser.AddLine(strings.TrimSuffix(line, "\r"))
if !ok {
continue
}
require.True(t, json.Valid([]byte(frame.Data)), "each downstream SSE frame must contain exactly one JSON document")
var event struct {
Type string `json:"type"`
}
require.NoError(t, json.Unmarshal([]byte(frame.Data), &event))
if frame.EventType != "" {
require.Equal(t, event.Type, frame.EventType)
}
eventTypes = append(eventTypes, event.Type)
}
if frame, ok := parser.Finish(); ok {
require.True(t, json.Valid([]byte(frame.Data)))
var event struct {
Type string `json:"type"`
}
require.NoError(t, json.Unmarshal([]byte(frame.Data), &event))
eventTypes = append(eventTypes, event.Type)
}
require.Equal(t, expectedTypes, eventTypes)
}
func openAIConcatenatedJSONTestEvents(t *testing.T) (string, string, string) {
t.Helper()
const javascriptErrorPosition = 68106
prefix := `{"type":"response.in_progress","response":{"id":"resp_large","status":"in_progress","instructions":"`
suffix := `"},"sequence_number":1}`
require.Less(t, len(prefix)+len(suffix), javascriptErrorPosition)
largeInProgress := prefix + strings.Repeat("x", javascriptErrorPosition-len(prefix)-len(suffix)) + suffix
outputItemAdded := `{"type":"response.output_item.added","output_index":0,"item":{"id":"msg_1","type":"message","role":"assistant","status":"in_progress","content":[]},"sequence_number":2}`
completed := `{"type":"response.completed","response":{"id":"resp_large","status":"completed","output":[],"usage":{"input_tokens":7,"output_tokens":9}},"sequence_number":3}`
require.Len(t, largeInProgress, javascriptErrorPosition)
require.True(t, json.Valid([]byte(largeInProgress)))
var decoded any
err := json.Unmarshal([]byte(largeInProgress+outputItemAdded), &decoded)
var syntaxErr *json.SyntaxError
require.ErrorAs(t, err, &syntaxErr)
require.Equal(t, int64(javascriptErrorPosition+1), syntaxErr.Offset)
return largeInProgress, outputItemAdded, completed
}
@@ -0,0 +1,115 @@
package service
import (
"bufio"
"bytes"
"encoding/json"
"io"
"strings"
)
const (
maxOpenAIConcatenatedJSONDocuments = 16
maxOpenAIConcatenatedJSONBytes = 16 * 1024 * 1024
)
// splitOpenAIConcatenatedJSONDocuments recognizes the narrow corruption shape
// produced when multiple complete Responses events arrive in one transport
// message. Other malformed payloads are left untouched for normal error paths.
func splitOpenAIConcatenatedJSONDocuments(payload []byte) ([][]byte, bool) {
payload = bytes.TrimSpace(payload)
if len(payload) == 0 || len(payload) > maxOpenAIConcatenatedJSONBytes || json.Valid(payload) {
return nil, false
}
decoder := json.NewDecoder(bytes.NewReader(payload))
documents := make([][]byte, 0, 2)
for {
var raw json.RawMessage
err := decoder.Decode(&raw)
if err != nil {
if err == io.EOF && len(documents) > 1 {
return documents, true
}
return nil, false
}
raw = bytes.TrimSpace(raw)
var envelope struct {
Type string `json:"type"`
}
if err := json.Unmarshal(raw, &envelope); err != nil {
return nil, false
}
eventType := strings.TrimSpace(envelope.Type)
if eventType == "" || strings.ContainsAny(eventType, "\r\n") {
return nil, false
}
if len(documents) == maxOpenAIConcatenatedJSONDocuments {
return nil, false
}
documents = append(documents, raw)
}
}
type openAISSEJSONDocumentScanner struct {
scanner *bufio.Scanner
pending []string
current string
}
func newOpenAISSEJSONDocumentScanner(scanner *bufio.Scanner) *openAISSEJSONDocumentScanner {
return &openAISSEJSONDocumentScanner{scanner: scanner}
}
func (s *openAISSEJSONDocumentScanner) Scan() bool {
if len(s.pending) > 0 {
s.current = s.pending[0]
s.pending = s.pending[1:]
return true
}
if s.scanner == nil || !s.scanner.Scan() {
return false
}
line := s.scanner.Text()
data, ok := extractOpenAISSEDataLine(line)
if !ok {
s.current = line
return true
}
if len(data) > maxOpenAIConcatenatedJSONBytes {
s.current = line
return true
}
documents, repaired := splitOpenAIConcatenatedJSONDocuments([]byte(data))
if !repaired {
s.current = line
return true
}
expanded := make([]string, 0, len(documents)*3)
for i, document := range documents {
if i > 0 {
var envelope struct {
Type string `json:"type"`
}
_ = json.Unmarshal(document, &envelope)
expanded = append(expanded, "event: "+strings.TrimSpace(envelope.Type))
}
expanded = append(expanded, "data: "+string(document), "")
}
s.current = expanded[0]
s.pending = expanded[1:]
return true
}
func (s *openAISSEJSONDocumentScanner) Text() string {
return s.current
}
func (s *openAISSEJSONDocumentScanner) Err() error {
if s.scanner == nil {
return nil
}
return s.scanner.Err()
}
@@ -430,9 +430,30 @@ func (s *OpenAIGatewayService) forwardOpenAIWSV2(
}
readTimeout := s.openAIWSReadTimeout()
var pendingJSONDocuments [][]byte
for {
message, readErr := lease.ReadMessageWithContextTimeout(ctx, readTimeout)
var message []byte
var readErr error
if len(pendingJSONDocuments) > 0 {
message = pendingJSONDocuments[0]
pendingJSONDocuments = pendingJSONDocuments[1:]
} else {
message, readErr = lease.ReadMessageWithContextTimeout(ctx, readTimeout)
if readErr == nil {
if documents, repaired := splitOpenAIConcatenatedJSONDocuments(message); repaired {
logOpenAIWSModeInfo(
"concatenated_json_repaired account_id=%d conn_id=%s documents=%d bytes=%d",
account.ID,
truncateOpenAIWSLogValue(connID, openAIWSIDValueMaxLen),
len(documents),
len(message),
)
message = documents[0]
pendingJSONDocuments = append(pendingJSONDocuments, documents[1:]...)
}
}
}
if readErr != nil {
lease.MarkBroken()
closeStatus, closeReason := summarizeOpenAIWSReadCloseError(readErr)
@@ -629,7 +650,10 @@ func (s *OpenAIGatewayService) forwardOpenAIWSV2(
}
if isTerminalEvent {
cleanExit = true
// A terminal event must be the final JSON document in its WS message.
// Ignore any tail for the completed client turn, but never reuse the
// ambiguous upstream connection for another request.
cleanExit = len(pendingJSONDocuments) == 0
break
}
}