1
0
mirror of https://github.com/searxng/searxng.git synced 2024-11-22 12:10:11 +01:00

[fix] searx.network.stream: fix memory leak

This commit is contained in:
Alexandre Flament 2021-09-28 15:26:34 +02:00
parent 2eab89b4ca
commit 29893cf816
3 changed files with 42 additions and 17 deletions

View File

@ -9,6 +9,7 @@ from types import MethodType
from timeit import default_timer
import httpx
import anyio
import h2.exceptions
from .network import get_network, initialize
@ -166,7 +167,7 @@ async def stream_chunk_to_queue(network, queue, method, url, **kwargs):
async for chunk in response.aiter_raw(65536):
if len(chunk) > 0:
queue.put(chunk)
except httpx.StreamClosed:
except (httpx.StreamClosed, anyio.ClosedResourceError):
# the response was queued before the exception.
# the exception was raised on aiter_raw.
# we do nothing here: in the finally block, None will be queued
@ -183,11 +184,35 @@ async def stream_chunk_to_queue(network, queue, method, url, **kwargs):
queue.put(None)
def _stream_generator(method, url, **kwargs):
queue = SimpleQueue()
network = get_context_network()
future = asyncio.run_coroutine_threadsafe(
stream_chunk_to_queue(network, queue, method, url, **kwargs),
get_loop()
)
# yield chunks
obj_or_exception = queue.get()
while obj_or_exception is not None:
if isinstance(obj_or_exception, Exception):
raise obj_or_exception
yield obj_or_exception
obj_or_exception = queue.get()
future.result()
def _close_response_method(self):
asyncio.run_coroutine_threadsafe(
self.aclose(),
get_loop()
)
# reach the end of _self.generator ( _stream_generator ) to an avoid memory leak.
# it makes sure that :
# * the httpx response is closed (see the stream_chunk_to_queue function)
# * to call future.result() in _stream_generator
for _ in self._generator: # pylint: disable=protected-access
continue
def stream(method, url, **kwargs):
@ -202,25 +227,15 @@ def stream(method, url, **kwargs):
httpx.Client.stream requires to write the httpx.HTTPTransport version of the
the httpx.AsyncHTTPTransport declared above.
"""
queue = SimpleQueue()
network = get_context_network()
future = asyncio.run_coroutine_threadsafe(
stream_chunk_to_queue(network, queue, method, url, **kwargs),
get_loop()
)
generator = _stream_generator(method, url, **kwargs)
# yield response
response = queue.get()
response = next(generator) # pylint: disable=stop-iteration-return
if isinstance(response, Exception):
raise response
response._generator = generator # pylint: disable=protected-access
response.close = MethodType(_close_response_method, response)
yield response
# yield chunks
chunk_or_exception = queue.get()
while chunk_or_exception is not None:
if isinstance(chunk_or_exception, Exception):
raise chunk_or_exception
yield chunk_or_exception
chunk_or_exception = queue.get()
future.result()
yield from generator

View File

@ -6,6 +6,7 @@ import asyncio
import logging
import threading
import anyio
import httpcore
import httpx
from httpx_socks import AsyncProxyTransport
@ -102,6 +103,9 @@ class AsyncProxyTransportFixed(AsyncProxyTransport):
# then each new request creates a new stream and raise the same WriteError
await close_connections_for_url(self, url)
raise e
except anyio.ClosedResourceError as e:
await close_connections_for_url(self, url)
raise httpx.CloseError from e
except httpx.RemoteProtocolError as e:
# in case of httpx.RemoteProtocolError: Server disconnected
await close_connections_for_url(self, url)
@ -130,6 +134,9 @@ class AsyncHTTPTransportFixed(httpx.AsyncHTTPTransport):
# then each new request creates a new stream and raise the same WriteError
await close_connections_for_url(self._pool, url)
raise e
except anyio.ClosedResourceError as e:
await close_connections_for_url(self._pool, url)
raise httpx.CloseError from e
except httpx.RemoteProtocolError as e:
# in case of httpx.RemoteProtocolError: Server disconnected
await close_connections_for_url(self._pool, url)

View File

@ -85,7 +85,10 @@ def _download_and_check_if_image(image_url: str) -> bool:
})
r = next(stream)
r.close()
is_image = r.headers["content-type"].startswith('image/')
if r.status_code == 200:
is_image = r.headers.get('content-type', '').startswith('image/')
else:
is_image = False
del r
del stream
return is_image