Tornado Example: Delegating an blocking task to a worker thread pool from an asynchronous request handler

gistfile1.py

from time import sleep
from tornado.httpserver import HTTPServer
from tornado.ioloop import IOLoop
from tornado.web import Application, asynchronous, RequestHandler
from multiprocessing.pool import ThreadPool

_workers = ThreadPool(10)

def run_background(func, callback, args=(), kwds={}):
    def _callback(result):
        IOLoop.instance().add_callback(lambda: callback(result))
    _workers.apply_async(func, args, kwds, _callback)

# blocking task like querying to MySQL
def blocking_task(n):
    sleep(n)
    return n

class Handler(RequestHandler):
    @asynchronous
    def get(self):
        run_background(blocking_task, self.on_complete, (10,))

    def on_complete(self, res):
        self.write("Test {0}<br/>".format(res))
        self.finish()

HTTPServer(Application([("/", Handler)],debug=True)).listen(8888)
IOLoop.instance().start()

@gwillem's example didn't work for me unless I changed asyncio.wrap_future into to_tornado_future. I made a decorator @blocking that works like @run_on_executor but uses the await syntax. Example:

import time
from functools import wraps
from concurrent.futures import ThreadPoolExecutor

import tornado
import tornado.ioloop
import tornado.gen
from tornado.platform.asyncio import to_tornado_future

io_loop = tornado.ioloop.IOLoop.current()

def blocking(method):
    """Wraps the method in an async method, and executes the function on `self.executor`."""
    @wraps(method)
    async def wrapper(self, *args, **kwargs):
        fut = self.executor.submit(method, self, *args, **kwargs)
        return await to_tornado_future(fut)
    return wrapper

class SomeClass:
    executor = ThreadPoolExecutor(4)

    @blocking
    def block_for_long_time(self):
        print("  > Starting with blocking...")
        time.sleep(1)
        print("  > Stopping with blocking...")
        return 3.1415

    async def test(self):
        """Emulates concurrent requests."""
        for i in range(4):
            io_loop.spawn_callback(self.spawn_me)
        print(">>> Waiting a while...")
        # Give the callbacks enough time to finish
        await tornado.gen.sleep(2)
        print(">>> Done.")

    async def spawn_me(self):
        """Emulates handling of a single request."""
        print(" >> spawned!")
        res = await self.block_for_long_time()
        print(" >> got", res)

io_loop.run_sync(SomeClass().test)

Expected output: (order of lines can differ for 'Stopping ...' and 'got...' lines):

>>> Waiting a while...
 >> spawned!
  > Starting with blocking...
 >> spawned!
  > Starting with blocking...
 >> spawned!
  > Starting with blocking...
 >> spawned!
  > Starting with blocking...
  > Stopping with blocking...
  > Stopping with blocking...
 >> got 3.1415
  > Stopping with blocking...
 >> got 3.1415
 >> got 3.1415
  > Stopping with blocking...
 >> got 3.1415
>>> Done.

Thank you @mclate. I have made a decorator out of this: https://gist.github.com/lucascosta/4ddd0afadfee75398536cb4125a8732b

Note for other travellers - ThreadPoolExecutor works with @run_on_executor but ProcessPoolExecutor does not, because the calling class is not serializeable as required by ProcessPoolExecutor. This confused me for a while.

seems like there are some changes in Tornado 5+ related to Futures, so decorator above didn't work for me. So I've modified it to:

def blocking(method):
    """
    Wraps the method in an async method, and executes the function on `self.executor`
    """

    @wraps(method)
    async def wrapper(self, *args, **kwargs):
        def work():
            return method(self, *args, **kwargs)
        return await io_loop.run_in_executor(self.executor, work)

    return wrapper