Merge branch 'router-for-me:main' into main

sdk/api/handlers/claude/code_handlers.go

@@ -146,10 +146,12 @@ func (h *ClaudeCodeAPIHandler) handleNonStreamingResponse(c *gin.Context, rawJSO
 	c.Header("Content-Type", "application/json")
 	alt := h.GetAlt(c)
 	cliCtx, cliCancel := h.GetContextWithCancel(h, c, context.Background())
+	stopKeepAlive := h.StartNonStreamingKeepAlive(c, cliCtx)
 
 	modelName := gjson.GetBytes(rawJSON, "model").String()
 
 	resp, errMsg := h.ExecuteWithAuthManager(cliCtx, h.HandlerType(), modelName, rawJSON, alt)
+	stopKeepAlive()
 	if errMsg != nil {
 		h.WriteErrorResponse(c, errMsg)
 		cliCancel(errMsg.Error)
@@ -159,13 +161,18 @@ func (h *ClaudeCodeAPIHandler) handleNonStreamingResponse(c *gin.Context, rawJSO
 	// Decompress gzipped responses - Claude API sometimes returns gzip without Content-Encoding header
 	// This fixes title generation and other non-streaming responses that arrive compressed
 	if len(resp) >= 2 && resp[0] == 0x1f && resp[1] == 0x8b {
-		gzReader, err := gzip.NewReader(bytes.NewReader(resp))
-		if err != nil {
-			log.Warnf("failed to decompress gzipped Claude response: %v", err)
+		gzReader, errGzip := gzip.NewReader(bytes.NewReader(resp))
+		if errGzip != nil {
+			log.Warnf("failed to decompress gzipped Claude response: %v", errGzip)
 		} else {
-			defer gzReader.Close()
-			if decompressed, err := io.ReadAll(gzReader); err != nil {
-				log.Warnf("failed to read decompressed Claude response: %v", err)
+			defer func() {
+				if errClose := gzReader.Close(); errClose != nil {
+					log.Warnf("failed to close Claude gzip reader: %v", errClose)
+				}
+			}()
+			decompressed, errRead := io.ReadAll(gzReader)
+			if errRead != nil {
+				log.Warnf("failed to read decompressed Claude response: %v", errRead)
 			} else {
 				resp = decompressed
 			}

sdk/api/handlers/gemini/gemini_handlers.go

@@ -336,7 +336,9 @@ func (h *GeminiAPIHandler) handleGenerateContent(c *gin.Context, modelName strin
 	c.Header("Content-Type", "application/json")
 	alt := h.GetAlt(c)
 	cliCtx, cliCancel := h.GetContextWithCancel(h, c, context.Background())
+	stopKeepAlive := h.StartNonStreamingKeepAlive(c, cliCtx)
 	resp, errMsg := h.ExecuteWithAuthManager(cliCtx, h.HandlerType(), modelName, rawJSON, alt)
+	stopKeepAlive()
 	if errMsg != nil {
 		h.WriteErrorResponse(c, errMsg)
 		cliCancel(errMsg.Error)

sdk/api/handlers/handlers.go

@@ -9,6 +9,7 @@ import (
 	"fmt"
 	"net/http"
 	"strings"
+	"sync"
 	"time"
 
 	"github.com/gin-gonic/gin"
@@ -113,6 +114,19 @@ func StreamingKeepAliveInterval(cfg *config.SDKConfig) time.Duration {
 	return time.Duration(seconds) * time.Second
 }
 
+// NonStreamingKeepAliveInterval returns the keep-alive interval for non-streaming responses.
+// Returning 0 disables keep-alives (default when unset).
+func NonStreamingKeepAliveInterval(cfg *config.SDKConfig) time.Duration {
+	seconds := 0
+	if cfg != nil {
+		seconds = cfg.NonStreamKeepAliveInterval
+	}
+	if seconds <= 0 {
+		return 0
+	}
+	return time.Duration(seconds) * time.Second
+}
+
 // StreamingBootstrapRetries returns how many times a streaming request may be retried before any bytes are sent.
 func StreamingBootstrapRetries(cfg *config.SDKConfig) int {
 	retries := defaultStreamingBootstrapRetries
@@ -294,6 +308,53 @@ func (h *BaseAPIHandler) GetContextWithCancel(handler interfaces.APIHandler, c *
 	}
 }
 
+// StartNonStreamingKeepAlive emits blank lines every 5 seconds while waiting for a non-streaming response.
+// It returns a stop function that must be called before writing the final response.
+func (h *BaseAPIHandler) StartNonStreamingKeepAlive(c *gin.Context, ctx context.Context) func() {
+	if h == nil || c == nil {
+		return func() {}
+	}
+	interval := NonStreamingKeepAliveInterval(h.Cfg)
+	if interval <= 0 {
+		return func() {}
+	}
+	flusher, ok := c.Writer.(http.Flusher)
+	if !ok {
+		return func() {}
+	}
+	if ctx == nil {
+		ctx = context.Background()
+	}
+
+	stopChan := make(chan struct{})
+	var stopOnce sync.Once
+	var wg sync.WaitGroup
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		ticker := time.NewTicker(interval)
+		defer ticker.Stop()
+		for {
+			select {
+			case <-stopChan:
+				return
+			case <-ctx.Done():
+				return
+			case <-ticker.C:
+				_, _ = c.Writer.Write([]byte("\n"))
+				flusher.Flush()
+			}
+		}
+	}()
+
+	return func() {
+		stopOnce.Do(func() {
+			close(stopChan)
+		})
+		wg.Wait()
+	}
+}
+
 // appendAPIResponse preserves any previously captured API response and appends new data.
 func appendAPIResponse(c *gin.Context, data []byte) {
 	if c == nil || len(data) == 0 {

sdk/api/handlers/handlers_stream_bootstrap_test.go

@@ -56,6 +56,14 @@ func (e *failOnceStreamExecutor) CountTokens(context.Context, *coreauth.Auth, co
 	return coreexecutor.Response{}, &coreauth.Error{Code: "not_implemented", Message: "CountTokens not implemented"}
 }
 
+func (e *failOnceStreamExecutor) HttpRequest(ctx context.Context, auth *coreauth.Auth, req *http.Request) (*http.Response, error) {
+	return nil, &coreauth.Error{
+		Code:       "not_implemented",
+		Message:    "HttpRequest not implemented",
+		HTTPStatus: http.StatusNotImplemented,
+	}
+}
+
 func (e *failOnceStreamExecutor) Calls() int {
 	e.mu.Lock()
 	defer e.mu.Unlock()

sdk/api/handlers/openai/openai_handlers.go

@@ -524,7 +524,9 @@ func (h *OpenAIAPIHandler) handleCompletionsNonStreamingResponse(c *gin.Context,
 
 	modelName := gjson.GetBytes(chatCompletionsJSON, "model").String()
 	cliCtx, cliCancel := h.GetContextWithCancel(h, c, context.Background())
+	stopKeepAlive := h.StartNonStreamingKeepAlive(c, cliCtx)
 	resp, errMsg := h.ExecuteWithAuthManager(cliCtx, h.HandlerType(), modelName, chatCompletionsJSON, "")
+	stopKeepAlive()
 	if errMsg != nil {
 		h.WriteErrorResponse(c, errMsg)
 		cliCancel(errMsg.Error)

sdk/api/handlers/openai/openai_responses_handlers.go

@@ -103,20 +103,17 @@ func (h *OpenAIResponsesAPIHandler) handleNonStreamingResponse(c *gin.Context, r
 
 	modelName := gjson.GetBytes(rawJSON, "model").String()
 	cliCtx, cliCancel := h.GetContextWithCancel(h, c, context.Background())
-	defer func() {
-		cliCancel()
-	}()
+	stopKeepAlive := h.StartNonStreamingKeepAlive(c, cliCtx)
 
 	resp, errMsg := h.ExecuteWithAuthManager(cliCtx, h.HandlerType(), modelName, rawJSON, "")
+	stopKeepAlive()
 	if errMsg != nil {
 		h.WriteErrorResponse(c, errMsg)
+		cliCancel(errMsg.Error)
 		return
 	}
 	_, _ = c.Writer.Write(resp)
-	return
-
-	// no legacy fallback
-
+	cliCancel()
 }
 
 // handleStreamingResponse handles streaming responses for Gemini models.