Curio is a coroutine-based library for concurrent Python systems programming. It provides standard programming abstractions such as as tasks, sockets, files, locks, and queues. It works on Unix and Windows. You'll find it to be familiar, small, fast, and fun
A Simple Example
Here is a concurrent TCP echo server directly implemented using sockets:
# echoserv.py from curio import run, spawn from curio.socket import * async def echo_server(address): sock = socket(AF_INET, SOCK_STREAM) sock.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1) sock.bind(address) sock.listen(5) print('Server listening at', address) async with sock: while True: client, addr = await sock.accept() await spawn(echo_client, client, addr, daemon=True) async def echo_client(client, addr): print('Connection from', addr) async with client: while True: data = await client.recv(100000) if not data: break await client.sendall(data) print('Connection closed') if __name__ == '__main__': run(echo_server, ('',25000))
If you've done network programming with threads, it looks almost identical. Moreover, it can handle thousands of clients even though no threads are being used inside.
Curio supports standard synchronization primitives (events, locks, recursive locks, semaphores, and condition variables), queues, subprocesses, as well as running tasks in threads and processes. The task model fully supports cancellation, task groups, timeouts, monitoring, and other features critical to writing reliable code.
Talks Related to Curio
Concepts related to Curio's design and general issues related to async programming have been described by Curio's creator in various conference talks and tutorials:
- Build Your Own Async, Workshop talk by David Beazley at PyCon India, 2019.
- The Other Async (Threads + Asyncio = Love), Keynote talk by David Beazley at PyGotham, 2017.
- Fear and Awaiting in Async, Keynote talk by David Beazley at PyOhio 2016.
- Topics of Interest (Async), Keynote talk by David Beazley at Python Brasil 2015.
- Python Concurrency from the Ground Up (LIVE), talk by David Beazley at PyCon 2015.
Questions and Answers
Q: What is the point of the Curio project?
A: Curio is async programming, reimagined as something smaller, faster, and easier to reason about. It is meant to be both educational and practical.
Q: Is Curio implemented using asyncio?
A: No. Curio is a standalone library directly created from low-level I/O primitives.
Q: Is Curio meant to be a clone of asyncio?
A: No. Although Curio provides a significant amount of overlapping functionality, the API is different. Compatibility with other libaries is not a goal.
Q: Can Curio interoperate with other event loops?
A: It depends on what you mean by the word "interoperate." Curio's preferred mechanism of communication with the external world is a queue. It is possible to communicate between Curio, threads, and other event loops using queues.
Q: How fast is Curio?
A: In rough benchmarking of the simple echo server shown here, Curio
runs about 90% faster than comparable code using coroutines in
asyncio. This was last measured on Linux using Python 3.7b3. Keep
in mind there is a lot more to overall application performance than
the performance of a simple echo server so your mileage might
vary. See the
examples/benchmark directory for various testing
Q: What is the future of Curio?
A: Curio should be viewed as a library of primitives related to concurrent systems programming. At this time, it is considered to be feature-complete--meaning that it is not expected to sprout new capabilities. It may be updated from time to time to fix bugs or support new versions of Python.
Q: Can I contribute?
A: Curio is not a community-based project seeking developers or maintainers. However, having it work reliably is important. If you've found a bug or have an idea for making it better, please feel file an issue.
The following people contributed ideas to early stages of the Curio project: Brett Cannon, Nathaniel Smith, Alexander Zhukov, Laura Dickinson, and Sandeep Gupta.
Curio is the creation of David Beazley (@dabeaz) who is also responsible for its maintenance. http://www.dabeaz.com
If you want to learn more about concurrent programming more generally, you should come take a course!