thimble

A thimble is a tool for playing with needle and thread safely. This library, thimble, wraps objects that have a blocking API with a non-blocking, Twisted-friendly Deferred API by means of thread pools.

Quick start

The main object you're interested in is thimble.Thimble. It takes a thread pool, a blocking object, and a list of method names that you would like to defer to the thread pool.

Here's our example blocking object:

>>> class Car(object):
...     wheels = 4
...     def drive_to(self, location):
...          # Assume the real implementation blocks.
...          return "driven to {0}".format(location)
>>> car = Car()

For demonstration purposes, we'll use a test doubles for the thread pool and reactor; in real code, you'll want to use the real thing.

>>> from thimble.test.util import FakeThreadPool, FakeReactor
>>> pool = FakeThreadPool()
>>> reactor = FakeReactor()

The pool hasn't been started yet. (We'll see why that matters in a minute.)

>>> pool.started
False

Create a Thimble:

>>> from thimble import Thimble
>>> car_thimble = Thimble(reactor, pool, car, ["drive_to"])

When accessing a method named in the list, you get an object wrapping it instead. Calling it returns a Deferred. Any arguments passed are passed verbatim to the wrapped method.

>>> def print_(s):
...     # can't use from __future__ import print_function because of a
...     # doctest limitation :-(
...     print s

>>> d = car_thimble.drive_to("work").addCallback(print_)
driven to work

This Deferred has already fired synchronously, because we're using a fake thread pool and reactor.

You can access other attributes of the wrapped object directly on the Thimble:

>>> car.wheels
4

If the thread pool that you pass to a Thimble hasn't been started yet when it first tries to use it, the Thimble will start it and schedule its shutdown. If you pass a thread pool that was already started, you are responsible for its shutdown. In this case, the thread pool was not started yet, so Thimble started it for you:

>>> pool.started
True

Shut down the reactor, and the reactor will ask the thread pool to stop right before shutting down itself.

>>> reactor.stop()
>>> pool.started
False

Using thimble in your code

Thread pools

You can choose to use the reactor thread pool, or create your own thread pool.

Using the reactor thread pool is potentially a bad idea. The reactor thread pool is shared between a lot of software by default, and is also used for DNS resolution. If your software blocks all the available threads in the pool (either by accident or because of a bug), that affects DNS resolution, which in turn can affect many other systems; if it doesn't affect those systems directly (because they, too, want to use the reactor thread pool).

It's probably most reliable to have a dedicated thread pool per application, for two reasons:

• The application probably knows best what a good size would be for the thread pool.
• It is an appropriate state to put the global state: if you were to put it in a library, different users of the library in the same process can end up tripping over each other.

Unfortunately, shared global state is pretty much how you do it:

from twisted.python.threadpool import ThreadPool
_the_thread_pool = _ThreadPool()

See the documentation for the ThreadPool class for more details; it allows you to specify a minimum and maximum number of threads. The default values are probably pretty reasonable.

Concurrency and thread safety

The number of threads you specify your thread pool to have is the number of threads that can try to access your object concurrently. it's up to you to make sure that the object is actually thread safe.

If you would like to provide a non-blocking API to an object that isn't thread safe, you can just limit the number of threads in the thread pool to 1, causing fully synchronized access. Keep in mind that attribute accesses for any attribute name that isn't in the blocking_methods list will still be performed synchronously by the calling thread.

Entry points

While subclassing Thimble may accidentally work, it is not recommended. I reserve the right to change the implementation in a way that might break that: for example, by introducing a metaclass.

It's probably better to write a small utility function that either constructs a new Thimble that uses a shared thread pool, or always returns the same thimble.

Changelog

Thimble uses SemVer.

v0.2.0

• Minor updates to the tox CI set up
• Upgraded dependencies

v0.1.1

• Added this changelog
• Spelling fixes
• Added a .gitignore
• Lots of documentation improvements

v0.1.0

Initial public release.