thredo

Thredo is Dead

Thread was a research topic related to the mixing of threads and async. I spoke about it at EuroPython 2018 (https://www.youtube.com/watch?v=U66KuyD3T0M). However, due to a lack of time and outside interest, the project has been abandoned. Feel free to look around however. -- Dave

Introduction

Thredo is threads on async. For the brave. Or the foolish. Only time will tell.

Note: Thredo requires the most up-to-date version of Curio--meaning the one that is checked out of the Curio GitHub repository at https://github.com/dabeaz/curio.

High Level Overview (The Big Idea)

Consider the following thread program involving a worker, a producer, and queue::

import threading
import queue
import time

def worker(q):
    while True:
        item = q.get()
        if item is None:
            break
        print('Got:', item)

def main():
    q = queue.Queue()
    t = threading.Thread(target=worker, args=(q,))
    t.start()
    for n in range(10):
        q.put(n)
        time.sleep(1)
    q.put(None)
    t.join()

main()

In this code, there are blocking operations such as q.get() and time.sleep(). This blocking is ultimately handled by the host operating system. Because of that, it is very difficult for Python to do anything related to the actual control or scheduling of threads. Once blocked, a thread stays blocked forever or until some event occurs that causes it to unblock.

Thredo re-envisions threads by redirecting all blocking operations to an async library. The code looks mostly the same except that you use the thredo module. For example:

import thredo

def worker(q):
    while True:
        item = q.get()
        if item is None:
            break
        print('Got:', item)

def main():
    q = thredo.Queue()
    t = thredo.spawn(worker, q)
    for n in range(10):
        q.put(n)
        thredo.sleep(1)
    q.put(None)
    t.join()

thredo.run(main)

The main reason you'd use thredo however is that it gives you extra features such as thread groups, cancellation, and more. For example, here's a more advanced version of the above code::

import thredo

def worker(q):
    try:
        while True:
            item = q.get()
            print('Got:', item)
            q.task_done()
    except thredo.ThreadCancelled:
        print('Worker cancelled')

def main():
    q = thredo.Queue()
    with thredo.ThreadGroup(wait=None) as workers:
        for n in range(4):
            workers.spawn(worker, q)

        for n in range(10):
            q.put(n)
            thredo.sleep(1)

        q.join()    

thredo.run(main)

Examples

The examples directory contains more examples of using thredo. The examples/euro directory contains coding samples from the EuroPython 2018 talk.

FAQ

Q: Is this going to turn into a full-fledged project?

A: It's too early to say.

Q: Isn't this sort of like using concurrent.futures?

A: No. concurrent.futures provides no mechanism for controlling threads and no mechanism for cancelling threads. Although it might appear like this is so, given that you can seemingly "cancel" a Future, this has no effect on thread execution. Once started, worked submitted to a thread pool in concurrent.futures runs to completion regardless of whether or not the associated Future is cancelled. Cancelling a future really only causes it to be abandoned if it hasn't yet started. If you cancel a thread in thredo, it is cleanly cancelled at the next blocking operation.

Q: Are Thredo Threads real Threads?

A: Yes. All threads are created using threading.Thread. They run concurrently according to the same rules as normal threads and can use all of the objects normally associated with threads (including synchronization primitives in threading). If you want to use things like thread-cancellation however, you need to make sure you use various objects provided in the thredo module. For example, waiting on a thredo.Lock can be cancelled whereas waiting on a threading.Lock can not.