Vasanko/sanic
2
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity
Files
d38fc1719127851afad9be18b94dc845b5049f01
sanic/sanic/server.py
Ashley Sommer 30e6a310f1 Pausable response streams (#1179) 
* This commit adds handlers for the asyncio/uvloop protocol callbacks for pause_writing and resume_writing.
These are needed for the correct functioning of built-in tcp flow-control provided by uvloop and asyncio.
This is somewhat of a breaking change, because the `write` function in user streaming callbacks now must be `await`ed.
This is necessary because it is possible now that the http protocol may be paused, and any calls to write may need to wait on an async event to be called to become unpaused.

Updated examples and tests to reflect this change.

This change does not apply to websocket connections. A change to websocket connections may be required to match this change.

* Fix a couple of PEP8 errors caused by previous rebase.

* update docs

add await syntax to response.write in response-streaming docs.

* remove commented out code from a test file
2018-08-18 18:12:13 -07:00

724 lines
26 KiB
Python
Raw Blame History

import asyncio
import os
import traceback
from functools import partial
from inspect import isawaitable
from multiprocessing import Process
from signal import (
SIGTERM, SIGINT, SIG_IGN,
signal as signal_func,
Signals
)
from socket import (
socket,
SOL_SOCKET,
SO_REUSEADDR,
)
from time import time
from httptools import HttpRequestParser
from httptools.parser.errors import HttpParserError
from multidict import CIMultiDict
try:
import uvloop
asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())
except ImportError:
pass
from sanic.log import logger, access_logger
from sanic.response import HTTPResponse
from sanic.request import Request
from sanic.exceptions import (
RequestTimeout, PayloadTooLarge, InvalidUsage, ServerError,
ServiceUnavailable)
current_time = None
class Signal:
stopped = False
class HttpProtocol(asyncio.Protocol):
__slots__ = (
# event loop, connection
'loop', 'transport', 'connections', 'signal',
# request params
'parser', 'request', 'url', 'headers',
# request config
'request_handler', 'request_timeout', 'response_timeout',
'keep_alive_timeout', 'request_max_size', 'request_class',
'is_request_stream', 'router',
# enable or disable access log purpose
'access_log',
# connection management
'_total_request_size', '_request_timeout_handler',
'_response_timeout_handler', '_keep_alive_timeout_handler',
'_last_request_time', '_last_response_time', '_is_stream_handler',
'_not_paused')
def __init__(self, *, loop, request_handler, error_handler,
signal=Signal(), connections=set(), request_timeout=60,
response_timeout=60, keep_alive_timeout=5,
request_max_size=None, request_class=None, access_log=True,
keep_alive=True, is_request_stream=False, router=None,
state=None, debug=False, **kwargs):
self.loop = loop
self.transport = None
self.request = None
self.parser = None
self.url = None
self.headers = None
self.router = router
self.signal = signal
self.access_log = access_log
self.connections = connections
self.request_handler = request_handler
self.error_handler = error_handler
self.request_timeout = request_timeout
self.response_timeout = response_timeout
self.keep_alive_timeout = keep_alive_timeout
self.request_max_size = request_max_size
self.request_class = request_class or Request
self.is_request_stream = is_request_stream
self._is_stream_handler = False
self._not_paused = asyncio.Event(loop=loop)
self._total_request_size = 0
self._request_timeout_handler = None
self._response_timeout_handler = None
self._keep_alive_timeout_handler = None
self._last_request_time = None
self._last_response_time = None
self._request_handler_task = None
self._request_stream_task = None
self._keep_alive = keep_alive
self._header_fragment = b''
self.state = state if state else {}
if 'requests_count' not in self.state:
self.state['requests_count'] = 0
self._debug = debug
self._not_paused.set()
@property
def keep_alive(self):
return (
self._keep_alive and
not self.signal.stopped and
self.parser.should_keep_alive())
# -------------------------------------------- #
# Connection
# -------------------------------------------- #
def connection_made(self, transport):
self.connections.add(self)
self._request_timeout_handler = self.loop.call_later(
self.request_timeout, self.request_timeout_callback)
self.transport = transport
self._last_request_time = current_time
def connection_lost(self, exc):
self.connections.discard(self)
if self._request_timeout_handler:
self._request_timeout_handler.cancel()
if self._response_timeout_handler:
self._response_timeout_handler.cancel()
if self._keep_alive_timeout_handler:
self._keep_alive_timeout_handler.cancel()
def pause_writing(self):
self._not_paused.clear()
def resume_writing(self):
self._not_paused.set()
def request_timeout_callback(self):
# See the docstring in the RequestTimeout exception, to see
# exactly what this timeout is checking for.
# Check if elapsed time since request initiated exceeds our
# configured maximum request timeout value
time_elapsed = current_time - self._last_request_time
if time_elapsed < self.request_timeout:
time_left = self.request_timeout - time_elapsed
self._request_timeout_handler = (
self.loop.call_later(time_left,
self.request_timeout_callback)
)
else:
if self._request_stream_task:
self._request_stream_task.cancel()
if self._request_handler_task:
self._request_handler_task.cancel()
try:
raise RequestTimeout('Request Timeout')
except RequestTimeout as exception:
self.write_error(exception)
def response_timeout_callback(self):
# Check if elapsed time since response was initiated exceeds our
# configured maximum request timeout value
time_elapsed = current_time - self._last_request_time
if time_elapsed < self.response_timeout:
time_left = self.response_timeout - time_elapsed
self._response_timeout_handler = (
self.loop.call_later(time_left,
self.response_timeout_callback)
)
else:
if self._request_stream_task:
self._request_stream_task.cancel()
if self._request_handler_task:
self._request_handler_task.cancel()
try:
raise ServiceUnavailable('Response Timeout')
except ServiceUnavailable as exception:
self.write_error(exception)
def keep_alive_timeout_callback(self):
# Check if elapsed time since last response exceeds our configured
# maximum keep alive timeout value
time_elapsed = current_time - self._last_response_time
if time_elapsed < self.keep_alive_timeout:
time_left = self.keep_alive_timeout - time_elapsed
self._keep_alive_timeout_handler = (
self.loop.call_later(time_left,
self.keep_alive_timeout_callback)
)
else:
logger.debug('KeepAlive Timeout. Closing connection.')
self.transport.close()
self.transport = None
# -------------------------------------------- #
# Parsing
# -------------------------------------------- #
def data_received(self, data):
# Check for the request itself getting too large and exceeding
# memory limits
self._total_request_size += len(data)
if self._total_request_size > self.request_max_size:
exception = PayloadTooLarge('Payload Too Large')
self.write_error(exception)
# Create parser if this is the first time we're receiving data
if self.parser is None:
assert self.request is None
self.headers = []
self.parser = HttpRequestParser(self)
# requests count
self.state['requests_count'] = self.state['requests_count'] + 1
# Parse request chunk or close connection
try:
self.parser.feed_data(data)
except HttpParserError:
message = 'Bad Request'
if self._debug:
message += '\n' + traceback.format_exc()
exception = InvalidUsage(message)
self.write_error(exception)
def on_url(self, url):
if not self.url:
self.url = url
else:
self.url += url
def on_header(self, name, value):
self._header_fragment += name
if value is not None:
if self._header_fragment == b'Content-Length' \
and int(value) > self.request_max_size:
exception = PayloadTooLarge('Payload Too Large')
self.write_error(exception)
try:
value = value.decode()
except UnicodeDecodeError:
value = value.decode('latin_1')
self.headers.append(
(self._header_fragment.decode().casefold(), value))
self._header_fragment = b''
def on_headers_complete(self):
self.request = self.request_class(
url_bytes=self.url,
headers=CIMultiDict(self.headers),
version=self.parser.get_http_version(),
method=self.parser.get_method().decode(),
transport=self.transport
)
# Remove any existing KeepAlive handler here,
# It will be recreated if required on the new request.
if self._keep_alive_timeout_handler:
self._keep_alive_timeout_handler.cancel()
self._keep_alive_timeout_handler = None
if self.is_request_stream:
self._is_stream_handler = self.router.is_stream_handler(
self.request)
if self._is_stream_handler:
self.request.stream = asyncio.Queue()
self.execute_request_handler()
def on_body(self, body):
if self.is_request_stream and self._is_stream_handler:
self._request_stream_task = self.loop.create_task(
self.request.stream.put(body))
return
self.request.body.append(body)
def on_message_complete(self):
# Entire request (headers and whole body) is received.
# We can cancel and remove the request timeout handler now.
if self._request_timeout_handler:
self._request_timeout_handler.cancel()
self._request_timeout_handler = None
if self.is_request_stream and self._is_stream_handler:
self._request_stream_task = self.loop.create_task(
self.request.stream.put(None))
return
self.request.body = b''.join(self.request.body)
self.execute_request_handler()
def execute_request_handler(self):
self._response_timeout_handler = self.loop.call_later(
self.response_timeout, self.response_timeout_callback)
self._last_request_time = current_time
self._request_handler_task = self.loop.create_task(
self.request_handler(
self.request,
self.write_response,
self.stream_response))
# -------------------------------------------- #
# Responding
# -------------------------------------------- #
def log_response(self, response):
if self.access_log:
extra = {
'status': getattr(response, 'status', 0),
}
if isinstance(response, HTTPResponse):
extra['byte'] = len(response.body)
else:
extra['byte'] = -1
extra['host'] = 'UNKNOWN'
if self.request is not None:
if self.request.ip:
extra['host'] = '{0}:{1}'.format(self.request.ip,
self.request.port)
extra['request'] = '{0} {1}'.format(self.request.method,
self.request.url)
else:
extra['request'] = 'nil'
access_logger.info('', extra=extra)
def write_response(self, response):
"""
Writes response content synchronously to the transport.
"""
if self._response_timeout_handler:
self._response_timeout_handler.cancel()
self._response_timeout_handler = None
try:
keep_alive = self.keep_alive
self.transport.write(
response.output(
self.request.version, keep_alive,
self.keep_alive_timeout))
self.log_response(response)
except AttributeError:
logger.error('Invalid response object for url %s, '
'Expected Type: HTTPResponse, Actual Type: %s',
self.url, type(response))
self.write_error(ServerError('Invalid response type'))
except RuntimeError:
if self._debug:
logger.error('Connection lost before response written @ %s',
self.request.ip)
keep_alive = False
except Exception as e:
self.bail_out(
"Writing response failed, connection closed {}".format(
repr(e)))
finally:
if not keep_alive:
self.transport.close()
self.transport = None
else:
self._keep_alive_timeout_handler = self.loop.call_later(
self.keep_alive_timeout,
self.keep_alive_timeout_callback)
self._last_response_time = current_time
self.cleanup()
async def drain(self):
await self._not_paused.wait()
def push_data(self, data):
self.transport.write(data)
async def stream_response(self, response):
"""
Streams a response to the client asynchronously. Attaches
the transport to the response so the response consumer can
write to the response as needed.
"""
if self._response_timeout_handler:
self._response_timeout_handler.cancel()
self._response_timeout_handler = None
try:
keep_alive = self.keep_alive
response.protocol = self
await response.stream(
self.request.version, keep_alive, self.keep_alive_timeout)
self.log_response(response)
except AttributeError:
logger.error('Invalid response object for url %s, '
'Expected Type: HTTPResponse, Actual Type: %s',
self.url, type(response))
self.write_error(ServerError('Invalid response type'))
except RuntimeError:
if self._debug:
logger.error('Connection lost before response written @ %s',
self.request.ip)
keep_alive = False
except Exception as e:
self.bail_out(
"Writing response failed, connection closed {}".format(
repr(e)))
finally:
if not keep_alive:
self.transport.close()
self.transport = None
else:
self._keep_alive_timeout_handler = self.loop.call_later(
self.keep_alive_timeout,
self.keep_alive_timeout_callback)
self._last_response_time = current_time
self.cleanup()
def write_error(self, exception):
# An error _is_ a response.
# Don't throw a response timeout, when a response _is_ given.
if self._response_timeout_handler:
self._response_timeout_handler.cancel()
self._response_timeout_handler = None
response = None
try:
response = self.error_handler.response(self.request, exception)
version = self.request.version if self.request else '1.1'
self.transport.write(response.output(version))
except RuntimeError:
if self._debug:
logger.error('Connection lost before error written @ %s',
self.request.ip if self.request else 'Unknown')
except Exception as e:
self.bail_out(
"Writing error failed, connection closed {}".format(
repr(e)), from_error=True
)
finally:
if self.parser and (self.keep_alive
or getattr(response, 'status', 0) == 408):
self.log_response(response)
try:
self.transport.close()
except AttributeError as e:
logger.debug('Connection lost before server could close it.')
def bail_out(self, message, from_error=False):
if from_error or self.transport.is_closing():
logger.error("Transport closed @ %s and exception "
"experienced during error handling",
self.transport.get_extra_info('peername'))
logger.debug('Exception:\n%s', traceback.format_exc())
else:
exception = ServerError(message)
self.write_error(exception)
logger.error(message)
def cleanup(self):
"""This is called when KeepAlive feature is used,
it resets the connection in order for it to be able
to handle receiving another request on the same connection."""
self.parser = None
self.request = None
self.url = None
self.headers = None
self._request_handler_task = None
self._request_stream_task = None
self._total_request_size = 0
self._is_stream_handler = False
def close_if_idle(self):
"""Close the connection if a request is not being sent or received
:return: boolean - True if closed, false if staying open
"""
if not self.parser:
self.transport.close()
return True
return False
def close(self):
"""
Force close the connection.
"""
if self.transport is not None:
self.transport.close()
self.transport = None
def update_current_time(loop):
"""Cache the current time, since it is needed at the end of every
keep-alive request to update the request timeout time
:param loop:
:return:
"""
global current_time
current_time = time()
loop.call_later(1, partial(update_current_time, loop))
def trigger_events(events, loop):
"""Trigger event callbacks (functions or async)
:param events: one or more sync or async functions to execute
:param loop: event loop
"""
for event in events:
result = event(loop)
if isawaitable(result):
loop.run_until_complete(result)
def serve(host, port, request_handler, error_handler, before_start=None,
after_start=None, before_stop=None, after_stop=None, debug=False,
request_timeout=60, response_timeout=60, keep_alive_timeout=5,
ssl=None, sock=None, request_max_size=None, reuse_port=False,
loop=None, protocol=HttpProtocol, backlog=100,
register_sys_signals=True, run_multiple=False, run_async=False,
connections=None, signal=Signal(), request_class=None,
access_log=True, keep_alive=True, is_request_stream=False,
router=None, websocket_max_size=None, websocket_max_queue=None,
websocket_read_limit=2 ** 16, websocket_write_limit=2 ** 16,
state=None, graceful_shutdown_timeout=15.0):
"""Start asynchronous HTTP Server on an individual process.
:param host: Address to host on
:param port: Port to host on
:param request_handler: Sanic request handler with middleware
:param error_handler: Sanic error handler with middleware
:param before_start: function to be executed before the server starts
listening. Takes arguments `app` instance and `loop`
:param after_start: function to be executed after the server starts
listening. Takes arguments `app` instance and `loop`
:param before_stop: function to be executed when a stop signal is
received before it is respected. Takes arguments
`app` instance and `loop`
:param after_stop: function to be executed when a stop signal is
received after it is respected. Takes arguments
`app` instance and `loop`
:param debug: enables debug output (slows server)
:param request_timeout: time in seconds
:param response_timeout: time in seconds
:param keep_alive_timeout: time in seconds
:param ssl: SSLContext
:param sock: Socket for the server to accept connections from
:param request_max_size: size in bytes, `None` for no limit
:param reuse_port: `True` for multiple workers
:param loop: asyncio compatible event loop
:param protocol: subclass of asyncio protocol class
:param request_class: Request class to use
:param access_log: disable/enable access log
:param websocket_max_size: enforces the maximum size for
incoming messages in bytes.
:param websocket_max_queue: sets the maximum length of the queue
that holds incoming messages.
:param websocket_read_limit: sets the high-water limit of the buffer for
incoming bytes, the low-water limit is half
the high-water limit.
:param websocket_write_limit: sets the high-water limit of the buffer for
outgoing bytes, the low-water limit is a
quarter of the high-water limit.
:param is_request_stream: disable/enable Request.stream
:param router: Router object
:return: Nothing
"""
if not run_async:
# create new event_loop after fork
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
if debug:
loop.set_debug(debug)
connections = connections if connections is not None else set()
server = partial(
protocol,
loop=loop,
connections=connections,
signal=signal,
request_handler=request_handler,
error_handler=error_handler,
request_timeout=request_timeout,
response_timeout=response_timeout,
keep_alive_timeout=keep_alive_timeout,
request_max_size=request_max_size,
request_class=request_class,
access_log=access_log,
keep_alive=keep_alive,
is_request_stream=is_request_stream,
router=router,
websocket_max_size=websocket_max_size,
websocket_max_queue=websocket_max_queue,
websocket_read_limit=websocket_read_limit,
websocket_write_limit=websocket_write_limit,
state=state,
debug=debug,
)
server_coroutine = loop.create_server(
server,
host,
port,
ssl=ssl,
reuse_port=reuse_port,
sock=sock,
backlog=backlog
)
# Instead of pulling time at the end of every request,
# pull it once per minute
loop.call_soon(partial(update_current_time, loop))
if run_async:
return server_coroutine
trigger_events(before_start, loop)
try:
http_server = loop.run_until_complete(server_coroutine)
except BaseException:
logger.exception("Unable to start server")
return
trigger_events(after_start, loop)
# Ignore SIGINT when run_multiple
if run_multiple:
signal_func(SIGINT, SIG_IGN)
# Register signals for graceful termination
if register_sys_signals:
_singals = (SIGTERM,) if run_multiple else (SIGINT, SIGTERM)
for _signal in _singals:
try:
loop.add_signal_handler(_signal, loop.stop)
except NotImplementedError:
logger.warning('Sanic tried to use loop.add_signal_handler '
'but it is not implemented on this platform.')
pid = os.getpid()
try:
logger.info('Starting worker [%s]', pid)
loop.run_forever()
finally:
logger.info("Stopping worker [%s]", pid)
# Run the on_stop function if provided
trigger_events(before_stop, loop)
# Wait for event loop to finish and all connections to drain
http_server.close()
loop.run_until_complete(http_server.wait_closed())
# Complete all tasks on the loop
signal.stopped = True
for connection in connections:
connection.close_if_idle()
# Gracefully shutdown timeout.
# We should provide graceful_shutdown_timeout,
# instead of letting connection hangs forever.
# Let's roughly calcucate time.
start_shutdown = 0
while connections and (start_shutdown < graceful_shutdown_timeout):
loop.run_until_complete(asyncio.sleep(0.1))
start_shutdown = start_shutdown + 0.1
# Force close non-idle connection after waiting for
# graceful_shutdown_timeout
coros = []
for conn in connections:
if hasattr(conn, "websocket") and conn.websocket:
coros.append(
conn.websocket.close_connection()
)
else:
conn.close()
_shutdown = asyncio.gather(*coros, loop=loop)
loop.run_until_complete(_shutdown)
trigger_events(after_stop, loop)
loop.close()
def serve_multiple(server_settings, workers):
"""Start multiple server processes simultaneously. Stop on interrupt
and terminate signals, and drain connections when complete.
:param server_settings: kw arguments to be passed to the serve function
:param workers: number of workers to launch
:param stop_event: if provided, is used as a stop signal
:return:
"""
server_settings['reuse_port'] = True
server_settings['run_multiple'] = True
# Handling when custom socket is not provided.
if server_settings.get('sock') is None:
sock = socket()
sock.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
sock.bind((server_settings['host'], server_settings['port']))
sock.set_inheritable(True)
server_settings['sock'] = sock
server_settings['host'] = None
server_settings['port'] = None
def sig_handler(signal, frame):
logger.info("Received signal %s. Shutting down.", Signals(signal).name)
for process in processes:
os.kill(process.pid, SIGTERM)
signal_func(SIGINT, lambda s, f: sig_handler(s, f))
signal_func(SIGTERM, lambda s, f: sig_handler(s, f))
processes = []
for _ in range(workers):
process = Process(target=serve, kwargs=server_settings)
process.daemon = True
process.start()
processes.append(process)
for process in processes:
process.join()
# the above processes will block this until they're stopped
for process in processes:
process.terminate()
server_settings.get('sock').close()
Reference in New Issue View Git Blame Copy Permalink