ZProc - Multi Processing on steroids
Project description
ZProc is a thought experiment, in making multi-tasking easier and accessible to everyone.
It focuses on automating various tasks related to message passing systems, for pythonistas. The eventual goal is to make Python a first-class language for doing multi-tasking.
So that you don't have to think about arcane details, like request-reply loops, reliable pub-sub, worker management, task distributiom, exception propagation, that kind of thing...
Behold, the power of ZProc:
import zproc
#
# define "atomic" operations
#
@zproc.atomic
def eat_cookie(snapshot):
snapshot["cookies"] -= 1
print("nom nom nom")
@zproc.atomic
def bake_cookie(snapshot):
snapshot["cookies"] += 1
print("Here's a cookie!")
#
# specify a process
#
def cookie_eater(ctx):
state = ctx.create_state({'ready': True})
# `start_time=0` accesses events from the very beginning.
for _ in state.get_when_change("cookies", start_time=0, count=5):
eat_cookie(state)
# boilerplate
ctx = zproc.Context(wait=True)
state = ctx.create_state({"cookies": 0})
# create a "ready" handle
ready = state.get_when_available("ready")
# spwan the process
proc = ctx.spawn(cookie_eater)
print(proc)
# wait for ready
next(ready)
# finally, bake some cookies.
for _ in range(5):
bake_cookie(state)
Result:
Process - pid: 10815 target: '__main__.cookie_eater' ppid: 10802 is_alive: True exitcode: None
Here's a cookie!
Here's a cookie!
nom nom nom
Here's a cookie!
nom nom nom
Here's a cookie!
nom nom nom
Here's a cookie!
nom nom nom
nom nom nom
The framework does message passing for you.
The state.get_when_change()
iterator supplies the cookie_eater
process with events asynchronously, as the state is updated.
The core idea
Message passing is cumbersome, error prone, and tedius -- because there is a lot of manual wiring involved.
The idea behind this project is to provide a pythonic API over widely accepted models in the message passing realm.
It started out with embracing shared state (but not shared memory).
Shared memory is frowned upon by almost everyone, due to the fact that memory is inherently dumb. Memory doesn't really care who's writing to it.
Shared state brings it's own perils, because its really hard to keep track of changes.
With the aid of the Actor Model, ZProc's state keeps a track of who's doing what. So much so, that it can act as a time-machine and give you state events from any defined time.
It then evolved to do handle exceptions across processes, failover, worker swarms, event sourcing, and other very useful features realated to multi-tasking.
The underlying architecture 100% message passing based, and hence scalable across many computers, with minimal modifications to the user code.
Well why not just X?
Each technology solution has it's place. Here is a rundown of why you would use ZProc over X.
-
X == asyncio
It's kinda in the name. Asyncio is strictly for I/O based concurrency. And then there's the fact that it implies extreme levels of refactoring.To quote Joe Armstrong,
I want one way to program computers, not many.
Install
$ pip install zproc
MIT License
Python 3.5+
Documentation
Wishlist
Here are some of the ideas that I wish were implemented in ZProc, but currently aren't.
- Redundant state-servers -- automatic selection/fallback.
- Green, Erlang-style processes. (requires Cpython internals)
- Process links, that automatically propagate errors across processes.
- Make namespaces horizontally scalable.
Features
-
🌠 Process management
- Remembers to cleanup processes when exiting, for general peace. (even when they're nested)
- Keeps a record of processes created using ZProc.
- 🔖
[ZProc] Cleaning up 'python' (13652)...
-
🌠 Worker/Process Maps
- Automatically manages worker processes, and delegates tasks to them.
- 🔖
-
🌠 Communicating sequential processes, at the core.
- No need for manual message passing.
- Watch for changes in state, without busy waiting.
- 🔖.
-
Deterministic state updates.
- Ships with a event messaging system that doesn't rely on flaky PUB/SUB.
- Go back in time, with a
TimeMachine
!.
-
Distributed, by default.
- Scalable to multiple computers, with minimal refactoring.
-
🌠 Atomic Operations
- Perform an arbitrary number of operations on state as a single, atomic operation.
- 🔖
-
🌠 Detailed, humane error logging for Proceeses.
[ZProc] Crash report: target: '__main__.p1' pid: 8959 ppid: 8944 Traceback (most recent call last): File "/home/dev/Projects/zproc/zproc/child.py", line 88, in main **self.target_kwargs File "/home/dev/Projects/zproc/zproc/child.py", line 65, in target_wrapper return self.target(*args, **kwargs) File "test.py", line 12, in p1 raise ValueError ValueError
Caveats
- The state only gets updated if you do it at the highest level.
This means that if you mutate objects deep down in the dict hirearchy, they wont be reflected in the global state. - The state should be pickle-able.
- It runs an extra Processes for managing the state.
They're fairly lightweight though, and shouldn't add too much weight to your application.
FAQ
-
Fast?
- Above all, ZProc is written for safety and the ease of use.
- However, since its written using ZMQ, it's plenty fast for most stuff.
-
Stable?
- Mostly. However, since it's still in its development stage, you should expect some API changes.
-
Production ready?
- Please don't use it in production right now.
-
Windows compatible?
- Probably?
Local development
# get the code
git clone https://github.com/pycampers/zproc.git
# install dependencies
cd zproc
pipenv install
pipenv install -d
# activate virtualenv
pipenv shell
# install zproc, and run tests
pip install -e .
pytest
Build documentation
cd docs
./build.sh
# open docs
google-chrome _build/index.html
# start a build loop
./build.sh loop
ZProc in the wild
Thanks
- Thanks to open logos for providing the wonderful ZProc logo.
- Thanks to pieter hintjens, for his work on the ZeroMQ library and for his amazing book.
- Thanks to tblib, ZProc can raise First-class Exceptions from the zproc server!
- Thanks to psutil, ZProc can handle nested procesess!
- Thanks to Kennith Rietz. His setup.py was used to host this project on pypi. Plus a lot of documentation structure is blatantly copied from his documentation on requests
ZProc is short for Zero Process.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file zproc-0.9.5.tar.gz
.
File metadata
- Download URL: zproc-0.9.5.tar.gz
- Upload date:
- Size: 28.2 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/1.13.0 pkginfo/1.5.0.1 requests/2.21.0 setuptools/40.8.0 requests-toolbelt/0.9.1 tqdm/4.31.1 CPython/3.6.8
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 9231798662da052def7ca01a89b4f8a0a258e5af6719149ea2225ae780cc1ab9 |
|
MD5 | 650c8978d05ad013ec4b390504554011 |
|
BLAKE2b-256 | 9e15193249928b0651604f486d3033a1a6d257a4967a7c22ef0fdfeb038f3d2f |
File details
Details for the file zproc-0.9.5-py2.py3-none-any.whl
.
File metadata
- Download URL: zproc-0.9.5-py2.py3-none-any.whl
- Upload date:
- Size: 35.0 kB
- Tags: Python 2, Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/1.13.0 pkginfo/1.5.0.1 requests/2.21.0 setuptools/40.8.0 requests-toolbelt/0.9.1 tqdm/4.31.1 CPython/3.6.8
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | d8757514e2bf1b3f2cf426272f038e8b3b798f6e0da7d9caf65e8fea18658338 |
|
MD5 | 700ae9e79e47fd48491d59e3aaa69e4a |
|
BLAKE2b-256 | ce6e82ef3dbfc49377e2cc260fff1a232377744d396b03e02f4118183c86a04e |