CODEWITHSUNDEEP
pythonmultithreadingpython-asyncio
Eran Moshe
Eran Moshe

Reputation: 3208

Reading files in parallel and parameterize class parameters

Suppose I have a class, And want to read few files from the disk in parallel, and parameterize class parameters. What is the most correct way to do it (and how)?

I thought about threading since it's only I/O actions.

Example of non-parallel implementation (1-Threading):

import pandas as pd


class DataManager(object):
    def __init__(self):
        self.a = None
        self.b = None
        self.c = None
        self.d = None
        self.e = None
        self.f = None

    def load_data(self):
        self.a = pd.read_csv('a.csv')
        self.b = pd.read_csv('b.csv')
        self.c = pd.read_csv('c.csv')
        self.d = pd.read_csv('d.csv')
        self.e = pd.read_csv('e.csv')
        self.f = pd.read_csv('f.csv')

if __name__ == '__main__':
    dm = DataManager()
    dm.load_data()
    # Main thread is waiting for load_data to finish.
    print("finished loading data")

Upvotes: 5

Views: 3213

Answers (2)

Aleksandr S
Aleksandr S

Reputation: 348

I/O operations are not CPU bounded in most cases so using multiple processes is an overkill. Using multiple threads can be good, but pb.read_csv not only reads the file but parses it what can be CPU bounded. I suggest you to read files from disk with asyncio as soon as it was initially made for this purpose. Here is the code to do so:

import asyncio
import aiofiles


async def read_file(file_name):
    async with aiofiles.open(file_name, mode='rb') as f:
        return await f.read()


def read_files_async(file_names: list) -> list:
    loop = asyncio.get_event_loop()
    return loop.run_until_complete(
        asyncio.gather(*[read_file(file_name) for file_name in file_names]))


if __name__ == '__main__':
    contents = read_files_async([f'files/file_{i}.csv' for i in range(10)])
    print(contents)

The function read_files_async returns the list of file contents (byte buffers), which you can pass to pd.read_csv.

I think optimization of files reading only should be enough but you can parse files contents in parallel with multiple processes (threads and async won't increase performance of parsing process):

import multiprocessing as mp

NUMBER_OF_CORES = 4
pool = mp.Pool(NUMBER_OF_CORES)
pool.map(pb.read_csv, contents)

You should set NUMBER_OF_CORES according to your machine spec.

Upvotes: 7

Samuel
Samuel

Reputation: 3801

Possible solution with Python3 ThreadPoolExecutor

    from concurrent.futures import ThreadPoolExecutor
    import queue
    import pandas as pd

    def load_data_worker(data_queue, file_name):
        data_queue.put(pd.read_csv(file_name))

    class DataManager(object):
        def __init__(self):
            self.data_queue = queue.Queue()
            self.data_arr = []

        def load_data(self):
            with ThreadPoolExecutor() as executor:
                executor.submit(load_data_woker, self.data_queue, 'a.csv')
                executor.submit(load_data_woker, self.data_queue, 'b.csv')
                # ... 
                executor.submit(load_data_woker, self.data_queue, 'f.csv')
           # dumping Queue of loaded data to array 
           self.data_arr = list(self.data_queue.queue)



    if __name__ == '__main__':
        dm = DataManager()
        dm.load_data()
        # Main thread is waiting for load_data to finish.
        print("finished loading data")

Upvotes: 1

Related Questions