kiwiPy

kiwiPy is a library that makes remote messaging using RabbitMQ (and possibly other message brokers) EASY. It was designed to support high-throughput workflows in big-data and computational science settings and is currently used by AiiDA for computational materials research around the world. That said, kiwiPy is entirely general and can be used anywhere where high-throughput and robust messaging are needed.

Here's what you get:

• RPC
• Broadcast (with filters)
• Task queue messages

Let's dive in, with some examples taken from the rmq tutorial. To see more detail head over to the documentation.

RPC

The client:

import kiwipy

with kiwipy.connect('amqp://localhost') as comm:
    # Send an RPC message
    print(" [x] Requesting fib(30)")
    response = comm.rpc_send('fib', 30).result()
    print((" [.] Got %r" % response))

(rmq_rpc_client.py source)

The server:

import threading
import kiwipy

def fib(comm, num):
    if num == 0:
        return 0
    if num == 1:
        return 1

    return fib(comm, num - 1) + fib(comm, num - 2)

with kiwipy.connect('amqp://127.0.0.1') as comm:
    # Register an RPC subscriber with the name 'fib'
    comm.add_rpc_subscriber(fib, 'fib')
    # Now wait indefinitely for fibonacci calls
    threading.Event().wait()

(rmq_rpc_server.py source)

Worker

Create a new task:

import sys
import kiwipy

message = ' '.join(sys.argv[1:]) or "Hello World!"

with rmq.connect('amqp://localhost') as comm:
    comm.task_send(message)

(rmq_new_task.py source)

And the worker:

import time
import threading
import kiwipy

print(' [*] Waiting for messages. To exit press CTRL+C')


def callback(_comm, task):
    print((" [x] Received %r" % task))
    time.sleep(task.count(b'.'))
    print(" [x] Done")


try:
    with kiwipy.connect('amqp://localhost') as comm:
        comm.add_task_subscriber(callback)
        threading.Event().wait()
except KeyboardInterrupt:
    pass

(rmq_worker.py source)

Citing

If you use kiwiPy directly or indirectly (e.g. by using AiiDA) then please cite:

Uhrin, M., & Huber, S. P. (2020). kiwiPy : Robust , high-volume , messaging for big-data and computational science workflows, 5, 4–6. http://doi.org/10.21105/joss.02351

This helps us to keep making community software.

Versioning

This software follows Semantic Versioning

Contributing

Want a new feature? Found a bug? Want to contribute more documentation or a translation perhaps?

Help is always welcome, get started with the contributing guide.

Development

This package utilises tox for unit test automation, and pre-commit for code style formatting and test automation.

To install these development dependencies:

pip install tox pre-commit

To run the unit tests:

tox

For the rmq tests you will require a running instance of RabbitMQ. One way to achieve this is using Docker and launching test/rmq/docker-compose.yml.

To run the pre-commit tests:

pre-commit run --all

To build the documentation:

tox -e docs-clean

Changes should be submitted as Pull Requests (PRs) to the develop branch.

Publishing Releases

1. Create a release PR/commit to the develop branch, updating kiwipy/version.py and CHANGELOG.md.
2. Fast-forward merge develop into the master branch
3. Create a release on GitHub (https://github.com/aiidateam/kiwipy/releases/new), pointing to the release commit on master, named v.X.Y.Z (identical to version in kiwipy/version.py)
4. This will trigger the continuous-deployment GitHub workflow which, if all tests pass, will publish the package to PyPi. Check this has successfully completed in the GitHub Actions tab (https://github.com/aiidateam/kiwipy/actions).

(if the release fails, delete the release and tag)