Recycling gunicorn worker (#800)

* add recycling feature to gunicorn worker

* add unit tests

* add more unit tests, and remove redundant trigger_events call

* fixed up unit tests

* make flake8 happy

* address feedbacks

* make flake8 happy

* add doc
This commit is contained in:
7 2017-06-22 13:26:50 -07:00 committed by Eli Uriegas
parent b5369e611c
commit f049a4ca67
4 changed files with 118 additions and 10 deletions

View File

@ -57,6 +57,12 @@ for Gunicorn `worker-class` argument:
gunicorn myapp:app --bind 0.0.0.0:1337 --worker-class sanic.worker.GunicornWorker
```
If your application suffers from memory leaks, you can configure Gunicorn to gracefully restart a worker
after it has processed a given number of requests. This can be a convenient way to help limit the effects
of the memory leak.
See the [Gunicorn Docs](http://docs.gunicorn.org/en/latest/settings.html#max-requests) for more information.
## Asynchronous support
This is suitable if you *need* to share the sanic process with other applications, in particular the `loop`.
However be advised that this method does not support using multiple processes, and is not the preferred way

View File

@ -75,7 +75,7 @@ class HttpProtocol(asyncio.Protocol):
signal=Signal(), connections=set(), request_timeout=60,
request_max_size=None, request_class=None, has_log=True,
keep_alive=True, is_request_stream=False, router=None,
**kwargs):
state=None, **kwargs):
self.loop = loop
self.transport = None
self.request = None
@ -99,6 +99,9 @@ class HttpProtocol(asyncio.Protocol):
self._request_handler_task = None
self._request_stream_task = None
self._keep_alive = keep_alive
self.state = state if state else {}
if 'requests_count' not in self.state:
self.state['requests_count'] = 0
@property
def keep_alive(self):
@ -154,6 +157,9 @@ class HttpProtocol(asyncio.Protocol):
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)
@ -389,7 +395,7 @@ def serve(host, port, request_handler, error_handler, before_start=None,
register_sys_signals=True, run_async=False, connections=None,
signal=Signal(), request_class=None, has_log=True, keep_alive=True,
is_request_stream=False, router=None, websocket_max_size=None,
websocket_max_queue=None):
websocket_max_queue=None, state=None):
"""Start asynchronous HTTP Server on an individual process.
:param host: Address to host on
@ -427,8 +433,6 @@ def serve(host, port, request_handler, error_handler, before_start=None,
if debug:
loop.set_debug(debug)
trigger_events(before_start, loop)
connections = connections if connections is not None else set()
server = partial(
protocol,
@ -445,7 +449,8 @@ def serve(host, port, request_handler, error_handler, before_start=None,
is_request_stream=is_request_stream,
router=router,
websocket_max_size=websocket_max_size,
websocket_max_queue=websocket_max_queue
websocket_max_queue=websocket_max_queue,
state=state
)
server_coroutine = loop.create_server(
@ -457,6 +462,7 @@ def serve(host, port, request_handler, error_handler, before_start=None,
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))
@ -464,6 +470,8 @@ def serve(host, port, request_handler, error_handler, before_start=None,
if run_async:
return server_coroutine
trigger_events(before_start, loop)
try:
http_server = loop.run_until_complete(server_coroutine)
except:

View File

@ -29,7 +29,7 @@ class GunicornWorker(base.Worker):
self.ssl_context = self._create_ssl_context(cfg)
else:
self.ssl_context = None
self.servers = []
self.servers = {}
self.connections = set()
self.exit_code = 0
self.signal = Signal()
@ -96,11 +96,16 @@ class GunicornWorker(base.Worker):
async def _run(self):
for sock in self.sockets:
self.servers.append(await serve(
state = dict(requests_count=0)
self._server_settings["host"] = None
self._server_settings["port"] = None
server = await serve(
sock=sock,
connections=self.connections,
state=state,
**self._server_settings
))
)
self.servers[server] = state
async def _check_alive(self):
# If our parent changed then we shut down.
@ -109,7 +114,15 @@ class GunicornWorker(base.Worker):
while self.alive:
self.notify()
if pid == os.getpid() and self.ppid != os.getppid():
req_count = sum(
self.servers[srv]["requests_count"] for srv in self.servers
)
if self.max_requests and req_count > self.max_requests:
self.alive = False
self.log.info(
"Max requests exceeded, shutting down: %s", self
)
elif pid == os.getpid() and self.ppid != os.getppid():
self.alive = False
self.log.info("Parent changed, shutting down: %s", self)
else:
@ -166,3 +179,4 @@ class GunicornWorker(base.Worker):
self.alive = False
self.exit_code = 1
self.cfg.worker_abort(self)
sys.exit(1)

View File

@ -3,7 +3,11 @@ import json
import shlex
import subprocess
import urllib.request
from unittest import mock
from sanic.worker import GunicornWorker
from sanic.app import Sanic
import asyncio
import logging
import pytest
@ -20,3 +24,79 @@ def test_gunicorn_worker(gunicorn_worker):
with urllib.request.urlopen('http://localhost:1337/') as f:
res = json.loads(f.read(100).decode())
assert res['test']
class GunicornTestWorker(GunicornWorker):
def __init__(self):
self.app = mock.Mock()
self.app.callable = Sanic("test_gunicorn_worker")
self.servers = {}
self.exit_code = 0
self.cfg = mock.Mock()
self.notify = mock.Mock()
@pytest.fixture
def worker():
return GunicornTestWorker()
def test_worker_init_process(worker):
with mock.patch('sanic.worker.asyncio') as mock_asyncio:
try:
worker.init_process()
except TypeError:
pass
assert mock_asyncio.get_event_loop.return_value.close.called
assert mock_asyncio.new_event_loop.called
assert mock_asyncio.set_event_loop.called
def test_worker_init_signals(worker):
worker.loop = mock.Mock()
worker.init_signals()
assert worker.loop.add_signal_handler.called
def test_handle_abort(worker):
with mock.patch('sanic.worker.sys') as mock_sys:
worker.handle_abort(object(), object())
assert not worker.alive
assert worker.exit_code == 1
mock_sys.exit.assert_called_with(1)
def test_handle_quit(worker):
worker.handle_quit(object(), object())
assert not worker.alive
assert worker.exit_code == 0
def test_run_max_requests_exceeded(worker):
loop = asyncio.new_event_loop()
worker.ppid = 1
worker.alive = True
sock = mock.Mock()
sock.cfg_addr = ('localhost', 8080)
worker.sockets = [sock]
worker.wsgi = mock.Mock()
worker.connections = set()
worker.log = mock.Mock()
worker.loop = loop
worker.servers = {
"server1": {"requests_count": 14},
"server2": {"requests_count": 15},
}
worker.max_requests = 10
worker._run = mock.Mock(wraps=asyncio.coroutine(lambda *a, **kw: None))
# exceeding request count
_runner = asyncio.ensure_future(worker._check_alive(), loop=loop)
loop.run_until_complete(_runner)
assert worker.alive == False
worker.notify.assert_called_with()
worker.log.info.assert_called_with("Max requests exceeded, shutting down: %s",
worker)