From 722c650bfdf5375fc9e2d444485156b97c085762 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Thu, 20 Aug 2026 19:47:05 -0700
Subject: [PATCH] test: cover generic HTTP streaming provider header forwarding

Add sync and async regression tests for the BaseLLMHTTPHandler streaming
path, which forwards provider response headers for the ~30 providers that
ride the generic handler and had no coverage. Also drop redundant setup
prose from the moonshot invoke test docstring.
---
 .../llm_translation/test_bedrock_moonshot.py  |  4 -
 .../custom_httpx/test_llm_http_handler.py     | 76 +++++++++++++++++++
 2 files changed, 76 insertions(+), 4 deletions(-)

diff --git a/tests/llm_translation/test_bedrock_moonshot.py b/tests/llm_translation/test_bedrock_moonshot.py
index 61364cf2caa..a82d1c6f029 100644
--- a/tests/llm_translation/test_bedrock_moonshot.py
+++ b/tests/llm_translation/test_bedrock_moonshot.py
@@ -209,10 +209,6 @@ class TestBedrockMoonshotInvoke(BaseLLMChatTest):
         endpoint with the messages body. Iteration of the stream itself is
         not exercised here — moonshot streaming delegates to the OpenAI
         parser and is covered by the OpenAI test suite.
-
-        Patch ``make_sync_call`` at its import site in
-        ``base_invoke_transformation`` so we observe the exact kwargs it is
-        called with at stream-wrapper construction time.
         """
         from litellm.utils import CustomStreamWrapper
 
diff --git a/tests/test_litellm/llms/custom_httpx/test_llm_http_handler.py b/tests/test_litellm/llms/custom_httpx/test_llm_http_handler.py
index fddd8d09dfc..b78b313e05c 100644
--- a/tests/test_litellm/llms/custom_httpx/test_llm_http_handler.py
+++ b/tests/test_litellm/llms/custom_httpx/test_llm_http_handler.py
@@ -2071,3 +2071,79 @@ async def test_anthropic_invalid_thinking_signature_retry_resigns_bedrock_reques
     retry_authorization = posts[1]["headers"]["Authorization"]
     assert retry_authorization.startswith("AWS4-HMAC-SHA256")
     assert retry_authorization != first_attempt_headers["Authorization"]
+
+
+_GENERIC_STREAM_SSE = (
+    b'data: {"id":"chatcmpl-1","object":"chat.completion.chunk","created":1,'
+    b'"model":"test-model","choices":[{"index":0,"delta":{"content":"hi"},'
+    b'"finish_reason":null}]}\n\n'
+    b"data: [DONE]\n\n"
+)
+
+
+def _generic_stream_upstream_response() -> httpx.Response:
+    return httpx.Response(
+        200,
+        headers={
+            "x-request-id": "generic-req-123",
+            "x-ratelimit-remaining-requests": "42",
+        },
+        content=_GENERIC_STREAM_SSE,
+        request=httpx.Request("POST", "https://fake-vllm.test/v1/chat/completions"),
+    )
+
+
+def test_generic_http_handler_sync_streaming_forwards_provider_response_headers():
+    """
+    Regression test for the generic BaseLLMHTTPHandler streaming path used by
+    ~30 providers (deepseek, groq, hosted_vllm, databricks, openrouter, ...).
+
+    The sync `completion()` streaming branch builds the CustomStreamWrapper from
+    `make_sync_call`, which returns the upstream response headers alongside the
+    stream. Those headers must reach the caller as `llm_provider-*` entries in
+    `_hidden_params["additional_headers"]`, which is what the proxy merges into
+    the client-facing response headers.
+    """
+    mock_client = Mock(spec=HTTPHandler)
+    mock_client.post = Mock(return_value=_generic_stream_upstream_response())
+
+    response = litellm.completion(
+        model="hosted_vllm/test-model",
+        messages=[{"role": "user", "content": "Hello"}],
+        api_base="https://fake-vllm.test/v1",
+        api_key="sk-test",
+        stream=True,
+        client=mock_client,
+    )
+
+    additional_headers = response._hidden_params["additional_headers"]
+    assert additional_headers["llm_provider-x-request-id"] == "generic-req-123"
+    assert additional_headers["llm_provider-x-ratelimit-remaining-requests"] == "42"
+
+    assert "".join([chunk.choices[0].delta.content or "" for chunk in response]) == "hi"
+
+
+@pytest.mark.asyncio
+async def test_generic_http_handler_async_streaming_forwards_provider_response_headers():
+    """
+    Companion to the sync test above for `acompletion_stream_function`, which
+    builds its CustomStreamWrapper from `make_async_call_stream_helper`.
+    """
+    mock_client = AsyncMock(spec=AsyncHTTPHandler)
+    mock_client.post = AsyncMock(return_value=_generic_stream_upstream_response())
+
+    response = await litellm.acompletion(
+        model="hosted_vllm/test-model",
+        messages=[{"role": "user", "content": "Hello"}],
+        api_base="https://fake-vllm.test/v1",
+        api_key="sk-test",
+        stream=True,
+        client=mock_client,
+    )
+
+    additional_headers = response._hidden_params["additional_headers"]
+    assert additional_headers["llm_provider-x-request-id"] == "generic-req-123"
+    assert additional_headers["llm_provider-x-ratelimit-remaining-requests"] == "42"
+
+    collected = [chunk async for chunk in response]
+    assert "".join([chunk.choices[0].delta.content or "" for chunk in collected]) == "hi"