From 3722645539f4cbf11028e69eea0643dee5cdc773 Mon Sep 17 00:00:00 2001
From: bnkaiteam <bnkaiteam@gmail.com>
Date: Thu, 8 Oct 2026 11:32:35 +0900
Subject: [PATCH] [Bugfix][Frontend] Accept successful 2xx responses for batch
 output uploads (#60507)

Signed-off-by: bnkaiteam <bnkaiteam@gmail.com>
Co-authored-by: Flora Feng <4florafeng@gmail.com>
---
 tests/entrypoints/launchers/test_run_batch.py | 29 ++++++++++++++-----
 vllm/entrypoints/launchers/run_batch.py       |  4 +--
 2 files changed, 24 insertions(+), 9 deletions(-)

diff --git a/tests/entrypoints/launchers/test_run_batch.py b/tests/entrypoints/launchers/test_run_batch.py
index f7a2505807..1376eb0645 100644
--- a/tests/entrypoints/launchers/test_run_batch.py
+++ b/tests/entrypoints/launchers/test_run_batch.py
@@ -1071,17 +1071,32 @@ def _make_aiohttp_put_session(status: int = 200, body_text: str = ""):
 
 
 @pytest.mark.asyncio
-async def test_upload_data_uploads_once_on_success():
-    """A successful upload must not be retried (regression guard)."""
-    session = _make_aiohttp_put_session(status=200)
-    with patch(
-        "vllm.entrypoints.launchers.run_batch.aiohttp.ClientSession",
-        return_value=session,
+@pytest.mark.parametrize("status", [200, 201, 204])
+@pytest.mark.parametrize("from_file", [False, True])
+async def test_upload_data_uploads_once_on_success(status, from_file, tmp_path):
+    """Successful PUT responses must not retry either upload path."""
+    session = _make_aiohttp_put_session(status=status)
+    data_or_file = "payload"
+    if from_file:
+        path = tmp_path / "output.jsonl"
+        path.write_text(data_or_file, encoding="utf-8")
+        data_or_file = str(path)
+
+    with (
+        patch(
+            "vllm.entrypoints.launchers.run_batch.aiohttp.ClientSession",
+            return_value=session,
+        ),
+        patch(
+            "vllm.entrypoints.launchers.run_batch.asyncio.sleep",
+            new_callable=AsyncMock,
+        ) as sleep,
     ):
         await upload_data(
-            "https://example.com/output.jsonl", "payload", from_file=False
+            "https://example.com/output.jsonl", data_or_file, from_file=from_file
         )
     assert session.put.call_count == 1
+    sleep.assert_not_awaited()
 
 
 @pytest.mark.asyncio
diff --git a/vllm/entrypoints/launchers/run_batch.py b/vllm/entrypoints/launchers/run_batch.py
index f54e87dafe..266592a731 100644
--- a/vllm/entrypoints/launchers/run_batch.py
+++ b/vllm/entrypoints/launchers/run_batch.py
@@ -385,7 +385,7 @@ async def upload_data(output_url: str, data_or_file: str, from_file: bool) -> No
                 if from_file:
                     with open(data_or_file, "rb") as file:
                         async with session.put(output_url, data=file) as response:
-                            if response.status != 200:
+                            if not 200 <= response.status < 300:
                                 error_text = await response.text()
                                 raise Exception(
                                     f"Failed to upload file.\n"
@@ -394,7 +394,7 @@ async def upload_data(output_url: str, data_or_file: str, from_file: bool) -> No
                                 )
                 else:
                     async with session.put(output_url, data=data_or_file) as response:
-                        if response.status != 200:
+                        if not 200 <= response.status < 300:
                             error_text = await response.text()
                             raise Exception(
                                 f"Failed to upload data.\n"