I\'ve been playing around a bit with the Tornado web server and have come to a point where I want to stop the web server (for example during unit testing). The following sim
If you need this behavior for unit testing, take a look at tornado.testing.AsyncTestCase.
By default, a new IOLoop is constructed for each test and is available as self.io_loop. This IOLoop should be used in the construction of HTTP clients/servers, etc. If the code being tested requires a global IOLoop, subclasses should override get_new_ioloop to return it.
If you need to start and stop an IOLoop for some other purpose and can't call ioloop.stop() from a callback for some reason, a multi-threaded implementation is possible. To avoid race conditions, however, you need to synchronize access to the ioloop, which is actually impossible. Something like the following will result in deadlock:
Thread 1:
with lock:
ioloop.start()
Thread 2:
with lock:
ioloop.stop()
because thread 1 will never release the lock (start() is blocking) and thread 2 will wait till the lock is released to stop the ioloop.
The only way to do it is for thread 2 to call ioloop.add_callback(ioloop.stop), which will call stop() on thread 1 in the event loop's next iteration. Rest assured, ioloop.add_callback() is thread-safe.
In case you do no want to bother with threads, you could catch a keyboard interrupt signal :
try:
tornado.ioloop.IOLoop.instance().start()
# signal : CTRL + BREAK on windows or CTRL + C on linux
except KeyboardInterrupt:
tornado.ioloop.IOLoop.instance().stop()
We want to use a multiprocessing.Process
with a tornado.ioloop.IOLoop
to work around the cPython GIL for performance and independency. To get access to the IOLoop we need to use Queue
to pass the shutdown signal through.
Here is a minimalistic example:
class Server(BokehServer)
def start(self, signal=None):
logger.info('Starting server on http://localhost:%d'
% (self.port))
if signal is not None:
def shutdown():
if not signal.empty():
self.stop()
tornado.ioloop.PeriodicCallback(shutdown, 1000).start()
BokehServer.start(self)
self.ioloop.start()
def stop(self, *args, **kwargs): # args important for signals
logger.info('Stopping server...')
BokehServer.stop(self)
self.ioloop.stop()
The Process
import multiprocessing as mp
import signal
from server import Server # noqa
class ServerProcess(mp.Process):
def __init__(self, *args, **kwargs):
self.server = Server(*args, **kwargs)
self.shutdown_signal = _mp.Queue(1)
mp.Process.__init__(self)
signal.signal(signal.SIGTERM, self.server.stop)
signal.signal(signal.SIGINT, self.server.stop)
def run(self):
self.server.start(signal=self.shutdown_signal)
def stop(self):
self.shutdown_signal.put(True)
if __name__ == '__main__':
p = ServerProcess()
p.start()
Cheers!
To stop the entire ioloop you simply invoke the ioloop.stop method when you have finished the unit test. (Remember that the only (documented) thread safe method is ioloop.add_callback, ie. if the unit tests is executed by another thread, you could wrap the stop invocation in a callback)
If its enough to stop the http web server you invoke the httpserver.stop() method
There is a problem with Zaar Hai's solution, namely that it leaves the socket open. The reason I was looking for a solution to stop Tornado is I'm running unit tests against my app server and I needed a way to start/stop the server between tests to have a clear state (empty session, etc.). By leaving the socket open, the second test always ran into an Address already in use
error. So I came up with the following:
import logging as log
from time import sleep
from threading import Thread
import tornado
from tornado.httpserver import HTTPServer
server = None
thread = None
def start_app():
def start():
global server
server = HTTPServer(create_app())
server.listen(TEST_PORT, TEST_HOST)
tornado.ioloop.IOLoop.instance().start()
global thread
thread = Thread(target=start)
thread.start()
# wait for the server to fully initialize
sleep(0.5)
def stop_app():
server.stop()
# silence StreamClosedError Tornado is throwing after it is stopped
log.getLogger().setLevel(log.FATAL)
ioloop = tornado.ioloop.IOLoop.instance()
ioloop.add_callback(ioloop.stop)
thread.join()
So the main idea here is to keep a reference to the HTTPServer
instance and call its stop()
method. And create_app()
just returns an Application
instance configured with handlers. Now you can use these methods in your unit tests like this:
class FoobarTest(unittest.TestCase):
def setUp(self):
start_app()
def tearDown(self):
stop_app()
def test_foobar(self):
# here the server is up and running, so you can make requests to it
pass
I just ran into this and found this issue myself, and using info from this thread came up with the following. I simply took my working stand alone Tornado code (copied from all the examples) and moved the actual starting code into a function. I then called the function as a threading thread. My case different as the threading call was done from my existing code where I just imported the startTornado and stopTornado routines.
The suggestion above seemed to work great, so I figured I would supply the missing example code. I tested this code under Linux on a FC16 system (and fixed my initial type-o).
import tornado.ioloop, tornado.web
class Handler(tornado.web.RequestHandler):
def get(self):
self.write("Hello, world")
application = tornado.web.Application([ (r"/", Handler) ])
def startTornado():
application.listen(8888)
tornado.ioloop.IOLoop.instance().start()
def stopTornado():
tornado.ioloop.IOLoop.instance().stop()
if __name__ == "__main__":
import time, threading
threading.Thread(target=startTornado).start()
print "Your web server will self destruct in 2 minutes"
time.sleep(120)
stopTornado()
Hope this helps the next person.