python多线程内存溢出--ThreadPoolExecutor内存溢出

ThreadPoolExecutor内存溢出

情景一：
在数据处理中，使用ThreadPoolExecutor（线程池）处理大量数据情况下，导致内存溢出
机器卡死挂掉；
场景模拟：

from concurrent.futures import ThreadPoolExecutor, as_completed
import time
from memory_profiler import profile
import queue
from line_profiler import LineProfiler
from functools import wraps


class BoundThreadPoolExecutor(ThreadPoolExecutor):
    """
    对ThreadPoolExecutor 进行重写，给队列设置边界
    """
    def __init__(self, qsize: int = None, *args, **kwargs):
        super(BoundThreadPoolExecutor, self).__init__(*args, **kwargs)
        self._work_queue = queue.Queue(qsize)

def timer(func):
    @wraps(func)
    def decorator(*args, **kwargs):
        func_return = func(*args, **kwargs)
        lp = LineProfiler()
        lp_wrap = lp(func)
        lp_wrap(*args, **kwargs)
        lp.print_stats()
        return func_return

    return decorator

def func(num):
    print(f"the {num} run...")
    time.sleep(0.5)
    return num*num

# @timer
@profile
def main():
    # with ThreadPoolExecutor(max_workers=2) as t:
    #     res = [t.submit(func, i) for i in range(100)]
    # pool = BoundThreadPoolExecutor(qsize=2, max_workers=2)
    pool = ThreadPoolExecutor(max_workers=2)
    for i in range(100):
        # func(i)
        pool.submit(func, i)
        print(pool._work_queue.qsize())
    pool.shutdown()

if __name__ == '__main__':
    main()

未对线程队列限制时，进程将所有对象添加到self._work_queue 中

重写ThreadPoolExecutor, 限制self._work_queue = queue.Queue(qsize)队列大小

结果对比

总结

存在内存溢出的情况，原因是ThreadPoolExecutor 线程池使用的是无边界队列，进程在队列中
添加对象时没有对空闲线程进行判断，导致内存消耗过多