This package provides an easy API for moving the work out of the tornado process / event loop.
Switch branches/tags
Nothing to show
Pull request Compare This branch is even with kmike:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.



This package provides an easy API for moving the work out of the tornado process / event loop.

Currently implemented methods are:

  • execute the code in another server's http hook (django implementation is included);
  • execute the code in a separate thread (thread pool is used);
  • dummy immediate execution.

API example:

from django.contrib.auth.models import User
from slacker import adisp
from slacker import Slacker
from slacker.workers import DjangoWorker

AsyncUser = Slacker(User, DjangoWorker())

def process_data():
    # all the django ORM is supported; the query will be executed
    # on remote end, this will not block the IOLoop

    users = yield AsyncUser.objects.filter(is_staff=True)[:5]
    print users

(pep-342 syntax and adisp library are optional, callback-style code is also supported)


pip install tornado-slacker

Slackers and workers

In order to execute some code in non-blocking manner:

  1. Create a Slacker (configured with the desired worker) for some python object:

    from slacker import Slacker
    from slacker.workers import ThreadWorker
    class Foo(object):
        # ...
    worker = ThreadWorker()
    AsyncFoo = Slacker(Foo, worker)
  2. build a query (you can access attributes, do calls and slicing):

    query = AsyncFoo('foo').do_blocking_operation(param1, param2)[0]
  3. execute the query:

    def callback(result):
        # ...

    or, using pep-342 style:

    from slacker import adisp
    def handler():
        result = yield query
        # ...


Slackers are special objects that are collecting operations (attribute access, calls, slicing) without actually executing them:

>>> from slacker import Slacker
>>> class Foo():
...     pass
>>> FooSlacker = Slacker(Foo)
__main__.Foo: [('hello',), ('world', (), {})]

>>> FooSlacker(name='me'), y=3)[:3]
__main__.Foo: [(None, (), {'name': 'me'}),
 ('world', (1,), {'y': 3}),
 (slice(None, 3, None), None)]

Callables arguments must be picklable. Slackers also provide a method to apply the collected operations to a base object.

Any picklable object (including top-level functions and classes) can be wrapped into Slacker, e.g.:

from slacker import adisp
from slacker import Slacker
from slacker.workers import ThreadWorker

def task(param1, param2):
    # do something blocking and io-bound
    return results

async_task = Slacker(task, ThreadWorker())

# pep-342-style
def process_data():
    results = yield async_task('foo', 'bar')
    print results

# callback style
def process_data2():
    async_task('foo', 'bar').proceed(on_result)

def on_result(results):
    print results

Python modules also can be Slackers:

import shutil
from slacker import Slacker
from slacker.workers import ThreadWorker

shutil_async = Slacker(shutil, ThreadWorker())
op = shutil_async.copy('file1.txt', 'file2.txt')


Workers are classes that decides how and where the work should be done:

  • slacker.workers.DummyWorker executes code in-place (this is blocking);

  • slacker.workers.ThreadWorker executes code in a thread from a thread pool;

  • slacker.workers.HttpWorker pickles the slacker, makes an async http request with this data to a given server hook and expects it to execute the code and return pickled results;


    IOLoop blocks on any CPU activity and making http requests plus unpickling the returned result can cause a significant overhead here. So if the query is fast (e.g. database primary key or index lookup, say 10ms) then it may be better not to use tornado-slacker and call the query in 'blocking' way: the overall blocking time may be less than with 'async' approach because of reduced computations amount.

    It is also wise to return as little as possible if HttpWorker is used.

  • slacker.workers.DjangoWorker is just a HttpWorker with default values for use with bundled django remote server hook implementation (slacker.django_backend).

    In order to enable django hook, include 'slacker.django_backend.urls' into and add SLACKER_SERVER option with server address to

    SLACKER_SERVER is '' by default so this should work for development server out of box.


    Do not expose django server hook to public, this is insecure! The best way is to configure additional server instance to listen some local port (e.g. bind it to the default address).


    Django's QuerySet arguments like Q, F objects, aggregate and annotate functions (e.g. Count) are picklable so tornado-slacker can handle them fine:

    AsyncAuthor = Slacker(Author, DjangoWorker())
    # ...
    qs = AsyncAuthor.objects.filter(
            Q(name='vasia') or Q(is_great=True)
    authors = yield qs

    Using slacker.Slacker is better than pickling queryset.query (as adviced at because this allows to pickle any ORM calls including ones that don't return QuerySets (

    yield AsyncUser.objects.create_superuser('foo')

    Moreover, slacker.Slacker adds transparent support for remote invocation of custom managers and model methods, returning just the model instance attributes, etc.

Parallel execution

Parallel task execution is supported by adisp library:

def _task1(param1, param2):
    # do something blocking
    return results

def _task2():
    # do something blocking
    return results

# worker can be reused
worker = ThreadWorker()
task1 = Slacker(_task1, worker)
task2 = Slacker(_task2, worker)

def process_data():
    # this will execute task1 and task2 in parallel
    # and return the result after all data is ready
    res1, res2 = yield task1('foo', 'bar'), task2()
    print res1, res2


this will fail with DjangoWorker and django development server because django development server is single-threaded


If you have any suggestions, bug reports or annoyances please report them to the issue tracker:

Source code:

Both hg and git pull requests are welcome!



Third-party software:

  • adisp (tornado adisp implementation is taken from brukva);
  • exception serialization utils are from billiard by Ask Solem.


The license is MIT.

Bundled adisp library uses Simplified BSD License.

slacker.serialization is under BSD License.