feat(frontend): session id plumbing into requests (#48048)

Signed-off-by: Karen Chung <karenc@nvidia.com>
This commit is contained in:
Karen Chung
2026-08-03 22:18:58 +00:00
committed by GitHub
parent e578de311c
commit f57123aa2d
39 changed files with 438 additions and 4 deletions
@@ -0,0 +1,97 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import pytest
from starlette.requests import Request
from vllm.entrypoints.generate.base.serving import GenerateBaseServing
from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest
from vllm.entrypoints.openai.completion.protocol import CompletionRequest
from vllm.entrypoints.openai.responses.protocol import ResponsesRequest
def _raw_request(headers: dict[str, str]) -> Request:
return Request(
{
"type": "http",
"method": "POST",
"path": "/v1/chat/completions",
"headers": [
(key.lower().encode("latin-1"), value.encode("latin-1"))
for key, value in headers.items()
],
}
)
@pytest.mark.parametrize(
"openai_request",
[
ChatCompletionRequest(
model="test-model",
messages=[{"role": "user", "content": "hi"}],
),
CompletionRequest(model="test-model", prompt="hi"),
ResponsesRequest(model="test-model", input="hi"),
],
)
def test_get_session_id_accepts_body_field(openai_request):
openai_request.session_id = "body-session"
session_id = GenerateBaseServing._get_session_id(
openai_request,
_raw_request({"X-Session-ID": "header-session"}),
)
assert session_id == "body-session"
def test_get_session_id_accepts_session_header():
request = CompletionRequest(model="test-model", prompt="hi")
session_id = GenerateBaseServing._get_session_id(
request,
_raw_request({"X-Session-ID": "header-session"}),
)
assert session_id == "header-session"
def test_get_session_id_ignores_correlation_header():
request = CompletionRequest(
model="test-model",
prompt="hi",
vllm_xargs={"session_id": "xargs-session"},
)
session_id = GenerateBaseServing._get_session_id(
request,
_raw_request({"X-Correlation-ID": "correlation-session"}),
)
assert session_id == "xargs-session"
def test_get_session_id_keeps_vllm_xargs_as_compatibility_fallback():
request = CompletionRequest(
model="test-model",
prompt="hi",
vllm_xargs={"session_id": "xargs-session"},
)
session_id = GenerateBaseServing._get_session_id(request, None)
assert session_id == "xargs-session"
def test_get_session_id_ignores_empty_and_non_string_values():
request = CompletionRequest(
model="test-model",
prompt="hi",
session_id="",
vllm_xargs={"session_id": 7},
)
session_id = GenerateBaseServing._get_session_id(request, None)
assert session_id is None
@@ -203,6 +203,32 @@ async def test_serve_tokens_skips_mm_cache_for_remote_engine_execution():
)
@pytest.mark.asyncio
async def test_serve_tokens_threads_session_id_header_to_engine():
engine = _mock_engine()
async def mock_generate(*args, **kwargs):
yield _make_request_output(
"req-1", token_ids=[10], finish_reason="stop", finished=True
)
engine.generate = MagicMock(side_effect=mock_generate)
serving = _build_serving_tokens(engine)
request = GenerateRequest(
token_ids=[1, 2, 3],
sampling_params=SamplingParams(max_tokens=1),
model=MODEL_NAME,
stream=False,
)
raw_request = MagicMock()
raw_request.headers = {"X-Session-ID": "header-session"}
await serving.serve_tokens(request, raw_request)
assert engine.generate.call_args.kwargs["session_id"] == "header-session"
@pytest.mark.asyncio
async def test_stream_basic():
"""Streaming returns SSE chunks with correct token_ids and ends with [DONE]."""
+16
View File
@@ -1,6 +1,8 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from copy import copy
from vllm import SamplingParams
from vllm.outputs import CompletionOutput
from vllm.sampling_params import RequestOutputKind
@@ -68,6 +70,20 @@ def test_parent_request_to_output_final_only() -> None:
)
def test_parallel_sampling_child_requests_preserve_session_id() -> None:
request = make_request(SamplingParams(n=2))
request.session_id = "session-1"
parent_request = ParentRequest(request)
for idx in range(parent_request.n):
request_id, child_params = parent_request.get_child_info(idx)
child_request = request if idx == parent_request.n - 1 else copy(request)
child_request.request_id = request_id
child_request.sampling_params = child_params
assert child_request.session_id == "session-1"
def make_request(sampling_params: SamplingParams) -> EngineCoreRequest:
return EngineCoreRequest(
request_id="parent_id",
+23 -1
View File
@@ -1,6 +1,9 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from vllm.v1.request import RequestStatus
from vllm import SamplingParams
from vllm.v1.engine import EngineCoreRequest
from vllm.v1.request import Request, RequestStatus
def test_request_status_fmt_str():
@@ -18,3 +21,22 @@ def test_request_status_fmt_str():
assert f"{RequestStatus.FINISHED_LENGTH_CAPPED}" == "FINISHED_LENGTH_CAPPED"
assert f"{RequestStatus.FINISHED_ABORTED}" == "FINISHED_ABORTED"
assert f"{RequestStatus.FINISHED_IGNORED}" == "FINISHED_IGNORED"
def test_request_copies_session_id_from_engine_core_request():
engine_request = EngineCoreRequest(
request_id="request-1",
prompt_token_ids=[1, 2, 3],
mm_features=None,
sampling_params=SamplingParams(max_tokens=1),
pooling_params=None,
arrival_time=0.0,
lora_request=None,
cache_salt=None,
data_parallel_rank=None,
session_id="session-1",
)
request = Request.from_engine_core_request(engine_request, block_hasher=None)
assert request.session_id == "session-1"