Skip to main content

Happy eyeballs and underlying scheduling algorithm in asyncio

Project description

Quick, what’s the situation?

To get all the benefits of Happy Eyeballs connection establishment algorithm, simply use async_stagger.open_connection like you would use asyncio.open_connection:

reader, writer = await async_stagger.open_connection('', 80)

Now your connections are more dual-stack friendly and will complete faster! A replacement for loop.create_connection is also provided.

The long version

What is Happy Eyeballs, and why should I use it?

Happy Eyeballs is an algorithm for establishing TCP connections to destinations specified by host names. It is described in RFC 6555 and RFC 8305. The primary benefit is that when host name resolution returns multiple addresses, and some of the address are unreachable, Happy Eyeballs will establish the connection much faster than conventional algorithms. For more information, check the Wikipedia article on Happy Eyeballs.

Python’s standard library provides several high-level methods of establishing TCP connections towards a host name: The socket module has socket.create_connection, and asyncio has loop.create_connection and asyncio.open_connection. By default, these methods have the same behavior when a host name resolves to several IP addresses: they try to connect to the first address in the list, and only after the attempt fails (which may take tens of seconds) will the second one be tried, and so on. In contrast, the Happy Eyeballs algorithm will start an attempt with the second IP address in parallel to the first one hasn’t completed after some time, typically around 300 milliseconds. As a result several attempts may be in flight at the same time, and whenever one of the attempts succeed, all other connections are cancelled, and the winning connection is used. This means a much shorter wait before one of the IP addresses connect successfully.

Happy Eyeballs is particularly important for dual-stack clients, when some hosts may have resolvable IPv6 addresses that are somehow unreachable.

Starting from Python 3.8, stock asyncio also supports Happy Eyeballs. See below for a comparison.

What does async_stagger has to offer?

async_stagger provides open_connection and create_connection with Happy Eyeballs support. They are mostly drop-in replacements for their asyncio counterparts, and support most existing arguments. (There are small differences: create_connection takes a loop argument instead of being a method on an event loop. Also, these two methods do not support the sock argument.) Another public coroutine create_connected_sock returns a connected socket.socket object. Check the documentation for details.

These methods implements many features specified in RFC 8305 Happy Eyeballs v2, which extends and obsoletes RFC 6555. In particular, asynchronous address resolution, destination address interleaving by family and staggered connection attempts are implemented.

Happy Eyeballs sounds great! I want to use similar logic somewhere else!

You’re in luck! async_stagger actually exposes the underlying scheduling logic as a reusable block: staggered_race. It can be use when:

  • There are several ways to achieve one goal. Some of the ways may fail, but you have to try it to find out.
  • Making attempts strictly in sequence is too slow.
  • You want to parallelize, but also don’t want to start the attempts all at the same time. Maybe you want to give preference to some of the attempts, so they should be started earlier and given more time to complete. Maybe you want to avoid straining the system with simultaneous attempts.
  • An attempt done half-way can be rolled back safely.

Where can I get it?

async_stagger requires Python 3.6 or later. (v0.2.0 onwards uses more new features in 3.6 such as async generators and async comprehensions, so it will probably require more than cosmetic changes to make it run on 3.5.) It does not have any external dependencies. Install it from PyPI the usual way:

pip install async-stagger

The documentation can be found here:

This project is under active development, and APIs may change in the future. Check out the Changelog in the documentation.

This project is licensed under the MIT license.

Python 3.8 Has Native Happy Eyeballs Now

I contributed an implementation of Happy Eyeballs to upstream asyncio, and it landed in Python 3.8: see the docs for details.

That implementation is essentially an early version of this package, so it lacks these features:

  • Async address resolution
  • Detailed exception report
  • The local_addrs argument (as opposed to local_addr)

Still, it should be sufficient for most scenarios, and it’s right there in the standard library.

Miscellaneous Remarks

Asynchronous address resolution is added in v0.2.1. With that, I feel that the package should be fairly feature-complete.

I have implemented Happy Eyeballs-like algorithms in some of my other projects, and this module reflects the things I have learned. However I have yet to eat my own dog food and actually import this module from those other projects. I would love to hear people’s experience using this module in real world conditions.

bpo-31861 talks about adding native aiter and anext functions either as builtins or to the operator module. Well, I want them NAO!!!one!!!eleventy!! So I borrowed the implementations from that bpo and put them in the aitertools submodule. I have only kept the one-argument forms; In particular, the two-argument iter function is so disparate from the one-argument version, that I don’t think they belong to the same function at all, and there really shouldn’t be a need for aiter to emulate that behavior.


The Happy Eyeballs scheduling algorithm implementation is inspired by the implementation in trio.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

async_stagger-0.3.1.tar.gz (23.4 kB view hashes)

Uploaded source

Built Distribution

async_stagger-0.3.1-py3-none-any.whl (25.3 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page