feat: sync deployed v1.0.0-rc.38-custom.42

Sync production source including routing, billing, Astra detection and public uptime. Update custom build instructions and license packaging.

relay/helper/stream_scanner.go

@@ -30,7 +30,8 @@ const (
 	// unconditional wg.Wait() in cleanup can always finish. Without it, a slow
 	// but connected client (full TCP buffer, no server WriteTimeout) could hang
 	// the handler forever.
-	streamWriteTimeout = 30 * time.Second
+	streamWriteTimeout     = 30 * time.Second
+	clientGoneDrainTimeout = 2 * time.Minute
 )
 
 func getScannerBufferSize() int {
@@ -40,9 +41,15 @@ func getScannerBufferSize() int {
 	return DefaultMaxScannerBufferSize
 }
 
-func NewStreamScanner(reader io.Reader) *bufio.Scanner {
+// NewStreamScanner shares relay scanner configuration. Callers buffering bounded
+// task state may additionally cap a line without increasing the configured limit.
+func NewStreamScanner(reader io.Reader, maxBytes ...int) *bufio.Scanner {
+	limit := getScannerBufferSize()
+	if len(maxBytes) > 0 && maxBytes[0] > 0 {
+		limit = min(limit, maxBytes[0])
+	}
 	scanner := bufio.NewScanner(reader)
-	scanner.Buffer(make([]byte, InitialScannerBufferSize), getScannerBufferSize())
+	scanner.Buffer(make([]byte, min(InitialScannerBufferSize, limit)), limit)
 	return scanner
 }
 
@@ -121,11 +128,11 @@ func StreamScannerHandler(c *gin.Context, resp *http.Response, info *relaycommon
 	logger.LogDebug(c, "streaming timeout seconds: %d", int64(streamingTimeout.Seconds()))
 	logger.LogDebug(c, "ping interval seconds: %d", int64(pingInterval.Seconds()))
 
-	cleanup := func() {
+	cleanup := func(closeUpstream bool) {
 		cleanupOnce.Do(func() {
 			cancel()
 			stop()
-			if resp.Body != nil {
+			if closeUpstream && resp.Body != nil {
 				_ = resp.Body.Close()
 			}
 
@@ -138,7 +145,9 @@ func StreamScannerHandler(c *gin.Context, resp *http.Response, info *relaycommon
 		})
 	}
 	// Ensure gin.Context is not returned to Gin's pool while any stream goroutine can still use it.
-	defer cleanup()
+	defer func() {
+		cleanup(true)
+	}()
 
 	scanner.Split(bufio.ScanLines)
 	copyCodexSSEHeaders(c, resp)
@@ -289,19 +298,44 @@ func StreamScannerHandler(c *gin.Context, resp *http.Response, info *relaycommon
 		info.StreamStatus.SetEndReason(relaycommon.StreamEndReasonEOF, nil)
 	})
 
+	// 客户端断开后仍读取上游终止事件中的 usage，避免上游已计费而
+	// NewAPI 无法结算。各写入函数会自行跳过已断开的客户端。
+	drainClientGone := func(err error) {
+		info.StreamStatus.SetClientGone(err)
+		logger.LogInfo(c, "client disconnected; draining upstream for billing usage")
+		drainTimer := time.NewTimer(clientGoneDrainTimeout)
+		select {
+		case <-stopChan:
+		case <-drainTimer.C:
+			logger.LogError(c, "client_gone upstream drain timeout; usage may be unavailable")
+		}
+		if !drainTimer.Stop() {
+			select {
+			case <-drainTimer.C:
+			default:
+			}
+		}
+	}
+
 	// 主循环等待完成或超时
 	select {
 	case <-ticker.C:
-		info.StreamStatus.SetEndReason(relaycommon.StreamEndReasonTimeout, nil)
+		if err := c.Request.Context().Err(); err != nil {
+			drainClientGone(err)
+		} else {
+			info.StreamStatus.SetEndReason(relaycommon.StreamEndReasonTimeout, nil)
+		}
 	case <-stopChan:
-		// EndReason already set by the goroutine that triggered stopChan
+		// A client disconnect can race with upstream [DONE]. Preserve it as the
+		// diagnostic reason while keeping fatal handler/scanner reasons intact.
+		if err := c.Request.Context().Err(); err != nil {
+			info.StreamStatus.SetClientGone(err)
+		}
 	case <-c.Request.Context().Done():
-		// 客户端断开：立即 cleanup 关闭上游 resp.Body，解除 scanner 阻塞并让上游停止生成，
-		// 避免为已放弃的请求继续消费上游 token。
-		info.StreamStatus.SetEndReason(relaycommon.StreamEndReasonClientGone, c.Request.Context().Err())
+		drainClientGone(c.Request.Context().Err())
 	}
 
-	cleanup()
+	cleanup(true)
 	if info.StreamStatus.IsNormalEnd() && !info.StreamStatus.HasErrors() {
 		logger.LogInfo(c, fmt.Sprintf("stream ended: %s", info.StreamStatus.Summary()))
 	} else {