fix: replay buffered success streams incrementally

scripts/test-gateway-e2e.mjs

@@ -299,11 +299,19 @@ function startFakeUpstream(port) {
           return;
         }
         if (parsed.stream) {
-          createSseResponse(res, [
-            `data: ${JSON.stringify({ type: "response.output_text.delta", delta: "hello", response_id: "resp_stream", thread_id: parsed.thread_id || "thread_stream", retry_attempt: reasoningAttempt })}\n\n`,
+          const deltaChunkCount = Number.isInteger(parsed.test_stream_delta_chunks) && parsed.test_stream_delta_chunks > 0
+            ? parsed.test_stream_delta_chunks
+            : 1;
+          const deltaTextBase = parsed.test_stream_delta_text || "hello";
+          const streamChunks = Array.from({ length: deltaChunkCount }, (_, index) => {
+            const deltaText = deltaChunkCount === 1 ? deltaTextBase : `${deltaTextBase}-${index + 1}`;
+            return `data: ${JSON.stringify({ type: "response.output_text.delta", delta: deltaText, response_id: "resp_stream", thread_id: parsed.thread_id || "thread_stream", retry_attempt: reasoningAttempt })}\n\n`;
+          });
+          streamChunks.push(
             `data: {"response":{"usage":{"output_tokens_details":{"reasoning_tokens":${reasoning}}}}}\n\n`,
             "data: [DONE]\n\n",
-          ], parsed.test_reasoning_response_delay_ms ?? parsed.test_stream_chunk_delay_ms ?? 20, {
+          );
+          createSseResponse(res, streamChunks, parsed.test_reasoning_response_delay_ms ?? parsed.test_stream_chunk_delay_ms ?? 20, {
             headers: reasoningAttempt
               ? { "x-upstream-reasoning-attempt": `${reasoningAttempt}` }
               : {},
@@ -430,40 +438,49 @@ function startGateway(configPath, logPath) {
   };
 }
 
-async function readSseUntilClose(url, requestBody) {
-  const response = await fetch(url, {
-    method: "POST",
-    headers: { "content-type": "application/json" },
-    body: JSON.stringify(requestBody),
-  });
+async function readSseUntilClose(url, requestBody) {
+  const startedAt = Date.now();
+  const response = await fetch(url, {
+    method: "POST",
+    headers: { "content-type": "application/json" },
+    body: JSON.stringify(requestBody),
+  });
 
-  const reader = response.body.getReader();
-  const decoder = new TextDecoder("utf8");
-  let text = "";
-  let closedByError = false;
-
-  while (true) {
-    try {
-      const { done, value } = await reader.read();
-      if (done) {
-        break;
-      }
-      text += decoder.decode(value, { stream: true });
-    } catch (error) {
-      closedByError = true;
-      text += `\n[[reader-error:${error?.name || "unknown"}]]`;
-      break;
+  const reader = response.body.getReader();
+  const decoder = new TextDecoder("utf8");
+  let text = "";
+  let closedByError = false;
+  let readCount = 0;
+  let firstChunkDelayMs = null;
+
+  while (true) {
+    try {
+      const { done, value } = await reader.read();
+      if (done) {
+        break;
+      }
+      readCount += 1;
+      if (firstChunkDelayMs === null) {
+        firstChunkDelayMs = Date.now() - startedAt;
+      }
+      text += decoder.decode(value, { stream: true });
+    } catch (error) {
+      closedByError = true;
+      text += `\n[[reader-error:${error?.name || "unknown"}]]`;
+      break;
     }
   }
 
   text += decoder.decode();
-  return {
-    status: response.status,
-    headers: response.headers,
-    text,
-    closedByError,
-  };
-}
+  return {
+    status: response.status,
+    headers: response.headers,
+    text,
+    closedByError,
+    readCount,
+    firstChunkDelayMs,
+  };
+}
 
 async function run() {
   assert(
@@ -1052,6 +1069,22 @@ async function run() {
       assert(okStream.status === 200, `${streamPath} 128 首状态异常: ${okStream.status}`);
       assert(okStream.text.includes("[DONE]"), `${streamPath} 流式 128 未完整结束`);
       assert(!okStream.closedByError, `${streamPath} 流式 128 不应异常断开`);
+
+      if (streamPath === "/responses" || streamPath === "/v1/responses") {
+        const replayedStream = await readSseUntilClose(
+          `http://127.0.0.1:${gatewayPort}${streamPath}`,
+          {
+            stream: true,
+            test_reasoning_tokens: 128,
+            test_stream_delta_chunks: 48,
+            test_stream_delta_text: "chunk",
+            test_stream_chunk_delay_ms: 2,
+          },
+        );
+        assert(replayedStream.status === 200, `${streamPath} 回放流首状态异常: ${replayedStream.status}`);
+        assert(replayedStream.text.includes("chunk-48"), `${streamPath} 回放流未保留尾部 delta`);
+        assert(replayedStream.readCount > 1, `${streamPath} 成功流不应退化为单块回放`);
+      }
     }
 
     const streamProgressPromise = fetch(`http://127.0.0.1:${gatewayPort}/responses`, {