Fix: stream AI chat answers under ASGI

This commit is contained in:
stumpylog committed 2026-11-17 07:42:51 -08:00
1 parent f68c093097
commit a690885a8a
2 files changed
+110 -9

No files matched your search

+81 -2
View File
@@ -4,14 +4,18 @@ from typing import TYPE_CHECKING
from unittest import mock
import pytest
from asgiref.sync import async_to_sync
from rest_framework import status
from rest_framework.test import APIClient
from rest_framework.test import APITestCase
from documents.views import _iterate_in_sync_thread
from paperless_testing.factories import UserFactory
from paperless_testing.permissions import grant_global
if TYPE_CHECKING:
from collections.abc import Generator
from django.contrib.auth.models import User
from pytest_mock import MockerFixture
@@ -62,7 +66,7 @@ class TestChatStreamingViewInputValidation(APITestCase):
),
mock.patch(
"paperless_ai.chat.stream_chat_with_documents",
return_value=iter(chunks),
return_value=(c for c in chunks),
),
):
resp = self.client.post(
@@ -74,7 +78,11 @@ class TestChatStreamingViewInputValidation(APITestCase):
assert resp.status_code == status.HTTP_200_OK
assert not resp.has_header("Content-Encoding")
assert list(resp.streaming_content) == [c.encode() for c in chunks]
async def read_parts() -> list[bytes]:
return [part async for part in resp.streaming_content]
assert async_to_sync(read_parts)() == [c.encode() for c in chunks]
def test_missing_question_is_rejected(self) -> None:
with mock.patch(
@@ -155,3 +163,74 @@ class TestChatStreamingViewUnrestrictedFlag:
assert (
mocked_stream_chat.call_args.kwargs["unrestricted"] is expected_unrestricted
)
class TestIterateInSyncThread:
@pytest.fixture
def events(self) -> list[str]:
return []
@pytest.fixture
def chunks(self, events: list[str]) -> Generator[str, None, None]:
def generate() -> Generator[str, None, None]:
try:
for chunk in ("first", "second"):
events.append(f"produced {chunk}")
yield chunk
finally:
events.append("closed")
return generate()
def test_yields_each_chunk_before_pulling_the_next(
self,
chunks: Generator[str, None, None],
events: list[str],
) -> None:
"""
GIVEN:
- A sync generator wrapped for async iteration
WHEN:
- The wrapper is consumed
THEN:
- Each chunk is delivered before the next one is generated,
rather than the whole generator being drained up front
"""
async def consume() -> None:
async for chunk in _iterate_in_sync_thread(chunks):
events.append(f"received {chunk}")
async_to_sync(consume)()
assert events == [
"produced first",
"received first",
"produced second",
"received second",
"closed",
]
def test_closing_early_closes_the_source_generator(
self,
chunks: Generator[str, None, None],
events: list[str],
) -> None:
"""
GIVEN:
- A sync generator wrapped for async iteration
WHEN:
- The consumer closes the wrapper after the first chunk, as
happens when the client disconnects
THEN:
- The source generator is closed, so its cleanup runs
"""
async def consume_one() -> None:
wrapper = _iterate_in_sync_thread(chunks)
await anext(wrapper)
await wrapper.aclose()
async_to_sync(consume_one)()
assert events == ["produced first", "closed"]