如何使用线程并行压缩迭代器?

Question

假设我有N 个生成器可以生成项目流gs = [..] # list of generators。

我可以轻松地将它们zip在一起，以gs：tuple_gen = zip(*gs)从每个相应的生成器中获取元组生成器。

这将按gs顺序调用每个g的next(g)，并在元组中收集结果。但是，如果每个项目的生产成本很高，我们可能希望在多个线程上并行化next(g)的工作。

如何实现执行此操作的pzip(..)？

Answer 1

你所要求的可以通过创建一个生成器来实现，该生成器从 ThreadPool 上的apply_async调用中产生结果。

仅供参考，我使用指定chunksize参数获得的pandas.read_csv迭代器对此方法进行了基准测试。我创建了一个 1M 行大小的 csv 文件的八个副本，并指定了 chunksize=100_000。

其中四个文件是使用您提供的顺序方法读取的，四个文件是使用下面的mt_gen函数读取的，使用四个线程的池：

• 单线程 ~ 3.68 秒
• 多线程 ~ 1.21 秒

不过，这并不意味着它会改善每个硬件和数据设置的结果。

import time
import threading
from multiprocessing.dummy import Pool  # dummy uses threads

def _load_sim(x = 10e6):
for _ in range(int(x)):
x -= 1
time.sleep(1)

def gen(start, stop):
for i in range(start, stop):
_load_sim()
print(f'{threading.current_thread().name} yielding {i}')
yield i

def multi_threaded(gens):
combi_g = mt_gen(gens)
for item in combi_g:
print(item)

def mt_gen(gens):
with Pool(N_WORKERS) as pool:
while True:
async_results = [pool.apply_async(next, args=(g,)) for g in gens]
try:
results = [r.get() for r in async_results]
except StopIteration:  # needed for Python 3.7+, PEP 479, bpo-32670
return
yield results

if __name__ == '__main__':
N_GENS = 10
N_WORKERS = 4
GEN_LENGTH = 3
gens = [gen(x * GEN_LENGTH, (x + 1) * GEN_LENGTH) for x in range(N_GENS)]
multi_threaded(gens)

输出：

Thread-1 yielding 0
Thread-2 yielding 3
Thread-4 yielding 6
Thread-3 yielding 9
Thread-1 yielding 12
Thread-2 yielding 15
Thread-4 yielding 18
Thread-3 yielding 21
Thread-1 yielding 24
Thread-2 yielding 27
[0, 3, 6, 9, 12, 15, 18, 21, 24, 27]
Thread-3 yielding 7
Thread-1 yielding 10
Thread-2 yielding 4
Thread-4 yielding 1
Thread-3 yielding 13
Thread-1 yielding 16
Thread-4 yielding 22
Thread-2 yielding 19
Thread-3 yielding 25
Thread-1 yielding 28
[1, 4, 7, 10, 13, 16, 19, 22, 25, 28]
Thread-1 yielding 8
Thread-4 yielding 2
Thread-3 yielding 11
Thread-2 yielding 5
Thread-1 yielding 14
Thread-4 yielding 17
Thread-3 yielding 20
Thread-2 yielding 23
Thread-1 yielding 26
Thread-4 yielding 29
[2, 5, 8, 11, 14, 17, 20, 23, 26, 29]
Process finished with exit code 0