badge.svg?branch=master&service=github)

Pytchfork simplifies working with python's multiprocessing package. By abstracting away the common boilerplate associated with forking processes and managing queues, pytchfork will allow you to write cleaner and more readable multiprocessing code.

#Usage ###Decorator You can easily mark methods to be run using multiple processes by invoking the pytchfork decorator:

from pytchfork import pytchfork
from multiprocessing import Queue

@pytchfork(3)
def do_work(queue):
  data = queue.get()
  process(data)

queue = Queue()
...
do_work(queue) # this call will fork 3 processes

Pytchfork can also manage queues for worker processes. Just provide the necessary references to the decorator and it will take care of polling the queue and passing data to the workers.

from pytchfork import pytchfork
from multiprocessing import Queue

@pytchfork(3, read_from=work_queue, write_to=done_queue, sentinel="DONE")
def process_data(data):
  processed_data = do_something(data)
  return processed_data

process_data() # this call will fork 3 processes that read from work_queue & write to done_queue.

####Redis Pytchfork processes can also be configured to read from and write to Redis instances (currently only Redis Lists are supported). To do so, simply pass the redis_uri and redis_port to the pytchfork decorator (In addition to string values Redis will use to key the work_queue and done_queue).

from pytchfork import pytchfork

@pytchfork(2, read_from="work_queue", write_to="done_queue", redis_uri='localhost', redis_port=6379)
def process_data(data):
  processed_data = do_something(data)
  return processed_data

process_data() # this will fork 2 processes that read from/write to a local redis instance

In the above snippet, these processes will run continuously as daemons. For smaller tasks with fixed amounts of input data, this might not be desirable.

To get the processes to exit upon completetion, pass the sentinel argument to the decorator. In order for this to work, the redis_work_queue must clearly mark the ending with N occurrences of the sentinel, where N is the desired number of processes.

Below is an example (verbosity for clarity):

from pytchfork import pytchfork
import redis

redis_client = redis.StrictRedis(host='localhost', port=6379)
fill_redis_with_work_tasks(redis_client, "work_queue")

num_procs = 2
sentinel = "DONE"
# mark ending here:
for x in range(0, num_procs):
  redis_client.lpush("work_queue", sentinel)

# provide the sentinel to the decorator
@pytchfork(num_procs, read_from="work_queue", write_to="done_queue", sentinel=sentinel, redis_uri=uri, redis_port=port)
def process_data(data):
  processed_data = do_something(data)
  return processed_data

process_data() # this will fork 2 processes that read/write to redis. Each process will
               # exit upon dequeueing a sentinel value from the redis work queue

For further reference on this, see the test_redis() method in tests/test_decorator.py.

###Context Manager You can also use the context manager to get hold of a multiprocessing.Pool object, without having to manage the lifecycle of the pool. I.e.:

from pytchfork import pytchfork
    ...
    with pytchfork(num_procs) as forked:
        res = forked.map_async(process_data, data, callback=callback)

This construct ensures that the worker processes will be closed, joined and terminated upon the completion of the code in the block.

##Installation Requirements: Python >= 2.7

$ pip install pytchfork

Contributing

Fork it ( https://github.com/shaunvxc/pytchfork/fork )
Create your feature branch (git checkout -b new-feature)
Commit your changes (git commit -am 'Add some feature')
Run the tests (make test)
Push change to the branch (git push origin new-feature)
Create a Pull Request