giambio/README.md

giambio - Asynchronous Python made easy (and friendly)

giambio is a Python framework that allows to perform I/O multiplexing and basically do more than one thing at once. This library implements what is known as a stackless mode of execution, or "green threads", though the latter term is misleading as no multithreading is involved (at least not by default).

Disclaimer

Right now this is no more than a toy implementation to help me understand how Python and async work, but it will (hopefully) become a production ready library soon

Also, this project was hugely inspired by the curio project and the trio project

Contributing

This is a relatively young project and it is looking for collaborators! It's not rocket science, but writing a proper framework like this implies some non-trivial issues that require proper and optimized solutions, so if you feel like you want to challenge yourself don't hesitate to contact me on Telegram or by E-mail

I'm no genius, I learned what I know from other people (reading articles, watching talks, failing to perform a certain task), so if you have enough experience with Python it's already enough!

Welcome in the world of coroutines

Libraries like giambio shine the most when it comes to performing asyncronous I/O (reading a socket, writing to a file, stuff like that).

The most common example of this is a network server, which just can't be handling one client at a time only.

One possible approach to achieve concurrency is to use threads, and despite their bad reputation in Python due to the GIL, they actually might be a good choice when it comes to I/O because, like every system call, it is performed out of control of the GIL as it does not execute any Python code. Problem is, if you have 10000 concurrent connections, you would need to build a system that can scale without your users being randomly disconnected when the server is overloaded, and that also does not need to spawn 10 thousands threads on your machine. On top of that, to avoid messing things up, you would need what is known as thread synchronization primitives and thread pools, to perform best. Such a big deal to run a web server, huh?

A library like giambio implements some low-level primitives (queues, semaphores, locks, events) and an Event Loop, that acts like a kernel, running behind the scenes: when a function needs to perform a blocking operation, it will trigger a "trap" that notifies the event loop that a certain task wants to perform an operation on a certain resource. The loop catches the trap and uses its associated metadata to perform the requested operation as soon as possible (which means that the operation might be completed instantly, but it's not guaranteed). But since most of the times doing I/O implies waiting for a resource to become available, the loop will also run other tasks while it's waiting for that to happen.

A deeper dive

giambio has been designed with simplicity in mind, so this README won't go deep in the explanation of what a coroutine is (you might want to check out this article if you want to learn more about all the implementation details and caveats about this). For the sake of this tutorial all you need to know is that a coroutine is a function defined with async def instead of the regular def, that a coroutine can call other coroutines, while synchronous functions can't, and that it enables the new Python 3.5 feature, which is the await keyword (basically an abstraction layer for yield).

Just to clarify things, giambio does not avoid the Global Interpreter Lock nor it performs any sort of multithreading or multiprocessing (at least by default). Remember that concurrency is not parallelism, concurrent tasks will switch back and forth and proceed with their calculations but won't be running independently like they would do if they were forked off to a process pool. That's why it is called concurrency, because multiple tasks concur for the same amount of resources. (Which is basically the same thing that happens inside your CPU at a much lower level, because processors run many more tasks than their actual number of cores)

If you read carefully, you might now wonder: "If a coroutine can call other coroutines, but synchronous functions cannot, how do I enter the async context in the first place?". This is done trough a special synchronous function (the start() method of an EventLoop object in our case) which can call asynchronous ones, that must be called from a synchronous context to avoid a horrible deadlock.

Let's code

Enough talking though, this is how a giambio based application looks like

import giambio
from giambio.socket import AsyncSocket
import socket
import logging

loop = giambio.EventLoop()

logging.basicConfig(level=20,
                    format="[%(levelname)s] %(asctime)s %(message)s",
                    datefmt='%d/%m/%Y %p')


async def make_srv(address: tuple):
    sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
    sock.bind(address)
    sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
    sock.listen(5)
    asock = loop.wrap_socket(sock)    # This creates a socket that can be read asynchronously
    logging.info(f"Echo server serving asynchronously at {address}")
    async with giambio.TaskManager(loop) as manager:
        while True:
            conn, addr = await asock.accept()
            logging.info(f"{addr} connected")
            task = manager.spawn(echo_server(conn, addr))   # This spawns a new task


async def echo_server(sock: AsyncSocket, addr: tuple):
    with sock:
        await sock.send_all(b"Welcome to the server pal!\n")
        while True:
            data = await sock.receive(1000)
            if not data:
                break
            to_send_back = data
            data = data.decode("utf-8").encode('unicode_escape')
            logging.info(f"Got: '{data.decode('utf-8')}' from {addr}")
            await sock.send_all(b"Got: " + to_send_back)
            logging.info(f"Echoed back '{data.decode('utf-8')}' to {addr}")
    logging.info(f"Connection from {addr} closed")


try:
    loop.start(make_srv, ('', 1501))
except KeyboardInterrupt:      # Because exceptions propagate
    print("Exiting...")

Explanation

Ok, let's explain this code line by line:

• First, we imported the required libraries
• Then, we created an EventLoop object
• For the sake of this tutorial, we built the "Hello world" of network servers, an echo server. An echo server always replies to the client with the same data that it got from it
• Here comes the real fun: In our make_srv function, which is an async function, we used a Python 3 feature that might look weird and unfamiliar to newcomers (especially if you are used to asyncio or similar frameworks). Giambio takes advantage of the async with context manager to perform its magic: The TaskManager object is an ideal space where all tasks are spawned and run until they are done.

The usage of the context manager ensures lots of cool things: For example, the context manager won't exit unless ALL the tasks inside it completed their execution (either cancelled, errored, or returned). This also means that because tasks are always joined automatically, you'll always get the return values of the coroutines and that exceptions will just work as expected.

You don't even need to be inside the with block to spawn tasks! All you need is a reference to the TaskManager object, so you can even pass it as a parameter to a function and spawn tasks from another coroutine and still get all the guarantees that giambio ensures.

Cancellation, exception propagation and return values

The only way to execute asynchronous code in giambio is trough a TaskManager (or one of its children classes) object used within an async with context manager. The TaskManager is an ideal "playground" where all asynchronous code runs and where the internal event loop of giambio can control its execution flow.

The key feature of this mechanism is that all tasks are always joined automatically: You'll never see a task running outside of the controlled context that giambio enforces.

Moreover, while most asynchronous frameworks out there would discard the return values from spawned tasks (in giambio "spawned" means that the coroutine was called using the spawn() or the schedule() method), here you'll never lose a single bit of information about your functions.

There are a few concepts to explain here, though:

• The term "task" refers to a coroutine executed trough the TaskManager's methods spawn() and schedule(), as well as one executed with await coro()

• Because of how the framework was designed, an exception in any of the task(s) inside the TaskManager will trigger the internal cancellation mechanism of giambio. All other running tasks are cancelled, read more below, and the exception(s) that caused the cancellation will be propagated inside the parent task as if you were running synchronous code

• The concept of cancellation is a bit tricky, because there is no real way to stop a coroutine from executing without actually raising an exception inside it. So when giambio needs to cancel a task, it just throws giambio.exceptions.CancelledError inside it and hopes for the best. This exception inherits from BaseException, which by convention means that it should not be catched. Doing so in giambio will likely break your code and make it explode spectacularly; If you really want to catch it to perform some sort of cleanup, be sure to re-raise it when done. In general, when writing an asynchronous function in giambio, you should always bear in mind that it might be cancelled at any time and handle that case accordingly.