python subclassing multiprocessing.Process

Question

I am new to python object oriented and I am rewriting my existing application as an object oriented version, because now developers are increasing and my code is becoming un-maintainable.

Normally I use multiprocessing queues but I found from this example http://www.doughellmann.com/PyMOTW/multiprocessing/basics.html that I can subclass multiprocessing.Process so I think it's a good idea and I wrote a class to test like this:

code:

from multiprocessing import Process
class Processor(Process):
    def return_name(self):
        return "Process %s" % self.name
    def run(self):
        return self.return_name()

processes = []


if __name__ == "__main__":

        for i in range(0,5):
                p=Processor()
                processes.append(p)
                p.start()
        for p in processes:
                p.join()

However I cannot get back the values, how can I use queues in this way?

EDIT: I want to get the return value and thinking where to put Queues().

Solution 1

Subclassing multiprocessing.Process:

However I cannot get back the values, how can I use queues in this way?

Process needs a Queue() to receive the results... An example of how to subclass multiprocessing.Process follows...

from multiprocessing import Process, Queue
class Processor(Process):

    def __init__(self, queue, idx, **kwargs):
        super(Processor, self).__init__()
        self.queue = queue
        self.idx = idx
        self.kwargs = kwargs

    def run(self):
        """Build some CPU-intensive tasks to run via multiprocessing here."""
        hash(self.kwargs) # Shameless usage of CPU for no gain...

        ## Return some information back through multiprocessing.Queue
        ## NOTE: self.name is an attribute of multiprocessing.Process
        self.queue.put("Process idx={0} is called '{1}'".format(self.idx, self.name))

if __name__ == "__main__":
    NUMBER_OF_PROCESSES = 5

    ## Create a list to hold running Processor object instances...
    processes = list()

    q = Queue()  # Build a single queue to send to all process objects...
    for i in range(0, NUMBER_OF_PROCESSES):
        p=Processor(queue=q, idx=i)
        p.start()
        processes.append(p)

    # Incorporating ideas from this answer, below...
    #    https://stackoverflow.com/a/42137966/667301
    [proc.join() for proc in processes]
    while not q.empty():
        print "RESULT: {0}".format(q.get())   # get results from the queue...

On my machine, this results in...

$ python test.py
RESULT: Process idx=0 is called 'Processor-1'
RESULT: Process idx=4 is called 'Processor-5'
RESULT: Process idx=3 is called 'Processor-4'
RESULT: Process idx=1 is called 'Processor-2'
RESULT: Process idx=2 is called 'Processor-3'
$

Using multiprocessing.Pool:

FWIW, one disadvantage I've found to subclassing multiprocessing.Process is that you can't leverage all the built-in goodness of multiprocessing.Pool; Pool gives you a very nice API if you don't need your producer and consumer code to talk to each other through a queue.

You can do a lot just with some creative return values... in the following example, I use a dict() to encapsulate input and output values from pool_job()...

from multiprocessing import Pool

def pool_job(input_val=0):
    # FYI, multiprocessing.Pool can't guarantee that it keeps inputs ordered correctly
    # dict format is {input: output}...
    return {'pool_job(input_val={0})'.format(input_val): int(input_val)*12}

pool = Pool(5)  # Use 5 multiprocessing processes to handle jobs...
results = pool.map(pool_job, xrange(0, 12)) # map xrange(0, 12) into pool_job()
print results

This results in:

[
    {'pool_job(input_val=0)': 0}, 
    {'pool_job(input_val=1)': 12}, 
    {'pool_job(input_val=2)': 24}, 
    {'pool_job(input_val=3)': 36}, 
    {'pool_job(input_val=4)': 48}, 
    {'pool_job(input_val=5)': 60}, 
    {'pool_job(input_val=6)': 72}, 
    {'pool_job(input_val=7)': 84}, 
    {'pool_job(input_val=8)': 96}, 
    {'pool_job(input_val=9)': 108}, 
    {'pool_job(input_val=10)': 120}, 
    {'pool_job(input_val=11)': 132}
]

Obviously there are plenty of other improvements to be made in pool_job(), such as error handling, but this illustrates the essentials. FYI, this answer provides another example of how to use multiprocessing.Pool.

Solution 2

The return value of Process.run doesn't go anywhere. You need to send them back to the parent process, e.g. using a multiprocessing.Queue (docs here).

Solution 3

Mike's answer is the best, but just for completeness I want to mention that I prefer harvesting the queue out of join contexts so the last bit would look like this:

[proc.join() for proc in processes] # 1. join

while not q.empty(): # 2. get the results
    print "RESULT: %s" % q.get()

Solution 4

Thanks a lot everyone.

Now heres how i got it done :)

In this example i use multiple queus as i do not want to communicate between each ohter but only with parent process.

from multiprocessing import Process,Queue
class Processor(Process):
    def __init__(self,queue):
        Process.__init__(self)
        self.que=queue
    def get_name(self):
        return "Process %s" % self.name
    def run(self):
        self.que.put(self.get_name())



if __name__ == "__main__":

        processes = []
        for i in range(0,5):
                p=Processor(Queue())
                processes.append(p)
                p.start()
        for p in processes:
                p.join()
                print p.que.get()

View more solutions

Comments

[Phyo Arkar Lwin]

So , in one of the method have to accept Queue object as parameter right?

[Phyo Arkar Lwin]

Please review my code and let me know what i can improve to be more pythonic / and better practice.

[Phyo Arkar Lwin]

Done! i created an init method accepting queues. this in-turn extends multiprocessing.Process to accept Queues directly :)

[Phyo Arkar Lwin]

THanks i will check. But i've read supers are dangerous especially multiple inheritance? is that true?

[Phyo Arkar Lwin]

Thanks for the correction. This code return self.queue.put(self.return_name()) returns a queue?

[Mike Pennington]

the return value from the Processor() call itself is irrelevant; you are passing the return value through the Queue() instance that you started the Processor() with

[Mike Pennington]

you need to be very careful when using super() with multiple inheritance. That said, I don't see multiple inheritance in the question you asked

[Thomas K]

You don't actually need a queue for each process. Multiple processes can put things into one queue without a problem. This is common when you have worker processes doing some computation, and putting output in a common result queue.

[Gerald]

with the super().__init__() method, do you pass in the target/arguments here, or in the self.run method?

[Verthais]

Does it really runs in parallel or does it get stuck on GIL while trying to access shared resource ( which in this case is our Queue ) ?