Improve concurrency on Windows (#286)

* Get future result in a callback

* POC with asyncio queue

* Clean up

* Update windows test wrapper

* WindowsProactorEventLoopPolicy on Windows

* Test setting future exception

* Update docstring

* Different errors depending on the platform

scrapy_playwright/_utils.py

@@ -1,5 +1,4 @@
 import asyncio
-import concurrent
 import logging
 import platform
 import threading
@@ -98,34 +97,66 @@ async def _get_header_value(
 
 if platform.system() == "Windows":
 
-    class _WindowsAdapter:
-        """Utility class to redirect coroutines to an asyncio event loop running
-        in a different thread. This allows to use a ProactorEventLoop, which is
-        supported by Playwright on Windows.
+    class _ThreadedLoopAdapter:
+        """Utility class to start an asyncio event loop in a new thread and redirect coroutines.
+        This allows to run Playwright in a different loop than the Scrapy crawler, allowing to
+        use ProactorEventLoop which is supported by Playwright on Windows.
         """
 
-        loop = None
-        thread = None
+        _loop: asyncio.AbstractEventLoop
+        _thread: threading.Thread
+        _coro_queue: asyncio.Queue = asyncio.Queue()
+        _stop_event: asyncio.Event = asyncio.Event()
 
         @classmethod
-        def get_event_loop(cls) -> asyncio.AbstractEventLoop:
-            if cls.thread is None:
-                if cls.loop is None:
-                    policy = asyncio.WindowsProactorEventLoopPolicy()  # type: ignore
-                    cls.loop = policy.new_event_loop()
-                    asyncio.set_event_loop(cls.loop)
-                if not cls.loop.is_running():
-                    cls.thread = threading.Thread(target=cls.loop.run_forever, daemon=True)
-                    cls.thread.start()
-                    logger.info("Started loop on separate thread: %s", cls.loop)
-            return cls.loop
+        async def _handle_coro(cls, coro, future) -> None:
+            try:
+                future.set_result(await coro)
+            except Exception as exc:
+                future.set_exception(exc)
 
         @classmethod
-        async def get_result(cls, coro) -> concurrent.futures.Future:
-            return asyncio.run_coroutine_threadsafe(coro=coro, loop=cls.get_event_loop()).result()
+        async def _process_queue(cls) -> None:
+            while not cls._stop_event.is_set():
+                coro, future = await cls._coro_queue.get()
+                asyncio.create_task(cls._handle_coro(coro, future))
+                cls._coro_queue.task_done()
 
-    def _deferred_from_coro(coro) -> Deferred:
-        return scrapy.utils.defer.deferred_from_coro(_WindowsAdapter.get_result(coro))
+        @classmethod
+        def _deferred_from_coro(cls, coro) -> Deferred:
+            future: asyncio.Future = asyncio.Future()
+            asyncio.run_coroutine_threadsafe(cls._coro_queue.put((coro, future)), cls._loop)
+            return scrapy.utils.defer.deferred_from_coro(future)
+
+        @classmethod
+        def start(cls) -> None:
+            policy = asyncio.WindowsProactorEventLoopPolicy()  # type: ignore[attr-defined]
+            cls._loop = policy.new_event_loop()
+            asyncio.set_event_loop(cls._loop)
+
+            cls._thread = threading.Thread(target=cls._loop.run_forever, daemon=True)
+            cls._thread.start()
+            logger.info("Started loop on separate thread: %s", cls._loop)
+
+            asyncio.run_coroutine_threadsafe(cls._process_queue(), cls._loop)
+
+        @classmethod
+        def stop(cls) -> None:
+            cls._stop_event.set()
+            asyncio.run_coroutine_threadsafe(cls._coro_queue.join(), cls._loop)
+            cls._loop.call_soon_threadsafe(cls._loop.stop)
+            cls._thread.join()
 
+    _deferred_from_coro = _ThreadedLoopAdapter._deferred_from_coro
 else:
+
+    class _ThreadedLoopAdapter:  # type: ignore[no-redef]
+        @classmethod
+        def start(cls) -> None:
+            pass
+
+        @classmethod
+        def stop(cls) -> None:
+            pass
+
     _deferred_from_coro = scrapy.utils.defer.deferred_from_coro

scrapy_playwright/handler.py

@@ -33,12 +33,13 @@
 from scrapy_playwright.headers import use_scrapy_headers
 from scrapy_playwright.page import PageMethod
 from scrapy_playwright._utils import (
+    _ThreadedLoopAdapter,
+    _deferred_from_coro,
     _encode_body,
     _get_header_value,
     _get_page_content,
     _is_safe_close_error,
     _maybe_await,
-    _deferred_from_coro,
 )
 
 
@@ -102,6 +103,7 @@ class ScrapyPlaywrightDownloadHandler(HTTPDownloadHandler):
 
     def __init__(self, crawler: Crawler) -> None:
         super().__init__(settings=crawler.settings, crawler=crawler)
+        _ThreadedLoopAdapter.start()
         if platform.system() != "Windows":
             verify_installed_reactor("twisted.internet.asyncioreactor.AsyncioSelectorReactor")
         crawler.signals.connect(self._engine_started, signals.engine_started)
@@ -293,6 +295,7 @@ def close(self) -> Deferred:
         logger.info("Closing download handler")
         yield super().close()
         yield _deferred_from_coro(self._close())
+        _ThreadedLoopAdapter.stop()
 
     async def _close(self) -> None:
         await asyncio.gather(*[ctx.context.close() for ctx in self.context_wrappers.values()])

tests/__init__.py

@@ -1,3 +1,4 @@
+import asyncio
 import inspect
 import logging
 import platform
@@ -13,17 +14,19 @@
 
 
 if platform.system() == "Windows":
-    from scrapy_playwright._utils import _WindowsAdapter
+    from scrapy_playwright._utils import _ThreadedLoopAdapter
 
     def allow_windows(test_method):
-        """Wrap tests with the _WindowsAdapter class on Windows."""
+        """Wrap tests with the _ThreadedLoopAdapter class on Windows."""
         if not inspect.iscoroutinefunction(test_method):
             raise RuntimeError(f"{test_method} must be an async def method")
 
         @wraps(test_method)
         async def wrapped(self, *args, **kwargs):
-            logger.debug("Calling _WindowsAdapter.get_result for %r", self)
-            await _WindowsAdapter.get_result(test_method(self, *args, **kwargs))
+            _ThreadedLoopAdapter.start()
+            coro = test_method(self, *args, **kwargs)
+            asyncio.run_coroutine_threadsafe(coro=coro, loop=_ThreadedLoopAdapter._loop).result()
+            _ThreadedLoopAdapter.stop()
 
         return wrapped
 

tests/tests_twisted/test_mixed_requests.py

@@ -15,11 +15,15 @@ class MixedRequestsTestCase(TestCase):
     '_download_request', which is a coroutine ('download_request' returns a Deferred).
     """
 
+    timeout_ms = 500
+
     @defer.inlineCallbacks
     def setUp(self):
         self.server = StaticMockServer()
         self.server.__enter__()
-        self.handler = ScrapyPlaywrightDownloadHandler.from_crawler(get_crawler())
+        self.handler = ScrapyPlaywrightDownloadHandler.from_crawler(
+            get_crawler(settings_dict={"PLAYWRIGHT_DEFAULT_NAVIGATION_TIMEOUT": self.timeout_ms})
+        )
         yield self.handler._engine_started()
 
     @defer.inlineCallbacks
@@ -29,26 +33,38 @@ def tearDown(self):
 
     @defer.inlineCallbacks
     def test_download_request(self):
-        def _test_regular(response, request):
+        def _check_regular(response, request):
             self.assertIsInstance(response, Response)
             self.assertEqual(response.css("a::text").getall(), ["Lorem Ipsum", "Infinite Scroll"])
             self.assertEqual(response.url, request.url)
             self.assertEqual(response.status, 200)
             self.assertNotIn("playwright", response.flags)
 
-        def _test_playwright(response, request):
+        def _check_playwright_ok(response, request):
             self.assertIsInstance(response, Response)
             self.assertEqual(response.css("a::text").getall(), ["Lorem Ipsum", "Infinite Scroll"])
             self.assertEqual(response.url, request.url)
             self.assertEqual(response.status, 200)
             self.assertIn("playwright", response.flags)
 
+        def _check_playwright_error(failure, url):
+            # different errors depending on the platform
+            self.assertTrue(
+                f"Page.goto: net::ERR_CONNECTION_REFUSED at {url}" in str(failure.value)
+                or f"Page.goto: Timeout {self.timeout_ms}ms exceeded" in str(failure.value)
+            )
+
         req1 = Request(self.server.urljoin("/index.html"))
         yield self.handler.download_request(req1, Spider("foo")).addCallback(
-            _test_regular, request=req1
+            _check_regular, request=req1
         )
 
         req2 = Request(self.server.urljoin("/index.html"), meta={"playwright": True})
         yield self.handler.download_request(req2, Spider("foo")).addCallback(
-            _test_playwright, request=req2
+            _check_playwright_ok, request=req2
+        )
+
+        req3 = Request("http://localhost:12345/asdf", meta={"playwright": True})
+        yield self.handler.download_request(req3, Spider("foo")).addErrback(
+            _check_playwright_error, url=req3.url
         )