在多核CPU下，同一进程下的多个线程可以并行运行吗

Posted 2023-04-30

　　CPU在某一个时间点上确实只能执行一个线程，但是多线程不是由于多核或者双核才叫多线程。
　　是由于，很多个线程在并行执行的时候，CPU根据一定的线程调度算法，频繁的进行线程切换，当正在执行的一个线程需要进行IO操作或者需要访问内存的时候，CPU完全可以放弃该线程，转而调度线程就绪队列上的其他线程，被放弃的线程则进入阻塞状态，IO操作或者访问内存操作结束之后，该线程可以进入线程就绪队列上。
　　人们通常意义上的多线程指的是，由于CPU根据一定的线程调度算法来切换线程，所以在一个时间段上，可以看做很多线程在并发执行。
　　其实还是在某一个时间点上只有一个线程在运行罢了。 参考技术A 同一个进程下的所有线程都只能在CPU同一个核下运行，同一进程下的多个线程在同一个核下轮流使用处理器，因为处理速度快，看起来是并行，实际上同一进程下的多线程是串行。
多核可以同时运行多个进程。 参考技术B 是的，多CPU正是为了多线程运行的。但是要注意线程间的同步。
纠正一个概念，是并发运行，不是并行运行。 参考技术C 现在的软件一般支持多线程的 参考技术D 那要看这个软件对多线程是否支持

Python3 系列之 并行编程

进程和线程

进程是程序运行的实例。一个进程里面可以包含多个线程，因此同一进程下的多个线程之间可以共享线程内的所有资源，它是操作系统动态运行的基本单元；每一个线程是进程下的一个实例，可以动态调度和独立运行，由于线程和进程有很多类似的特点，因此，线程又被称为轻量级的进程。线程的运行在进程之下，进程的存在依赖于线程；

开胃菜

基于 Python3 创建一个简单的进程示例

from threading import Thread
from time import sleep


class CookBook(Thread):
    def __init__(self):
        Thread.__init__(self)
        self.message = "Hello Parallel Python CookBook!!\n"

    def print_message(self):
        print(self.message)

    def run(self):
        print("Thread Starting\n")
        x = 0
        while x < 10:
            self.print_message()
            sleep(2)
            x += 1
        print("Thread Ended!\n")


print("Process Started")
hello_python = CookBook()

hello_python.start()
print("Process Ended")

需要注意的是，永远不要让线程在后台默默执行，当其执行完毕后要及时释放资源。

基于线程的并行

多线程编程一般使用共享内存空间进行线程间的通信，这就使管理内存空间成为多线程编程的关键。Python 通过标准库 threading 模块来管理线程，具有以下的组件：

• 线程对象
• Lock 对象
• RLock 对象
• 信号对象
• 条件对象
• 事件对象

定义一个线程

基本语法

示例代码如下所示

import threading


def function(i):
    print("function called by thread: 0".format(i))
    return


threads = []
for i in range(5):
    t = threading.Thread(target=function, args=(i,))
    threads.append(t)
    t.start()

lambda t, threads: t.join()

需要注意的是，线程创建后并不会自动运行，需要主动调用 start() 方法来启动线程，join() 会让调用它的线程被阻塞直到执行结束。（PS：可通过调用 t.setDaemon(True) 使其为后台线程避免主线程被阻塞）

线程定位

示例代码如下所示

import threading
import time


def first_function():
    print("0 is starting".format(threading.currentThread().getName()))
    time.sleep(2)
    print("0 is Exiting".format(threading.currentThread().getName()))


def second_function():
    print("0 is starting".format(threading.currentThread().getName()))
    time.sleep(2)
    print("0 is Exiting".format(threading.currentThread().getName()))


def third_function():
    print("0 is starting".format(threading.currentThread().getName()))
    time.sleep(2)
    print("0 is Exiting".format(threading.currentThread().getName()))

if __name__ == "__main__":
    t1 = threading.Thread(target=first_function,name="first")
    t2 = threading.Thread(target=second_function,name="second")
    t3 = threading.Thread(target=third_function,name="third")

    t1.start()
    t2.start()
    t3.start()
    t1.join()
    t2.join()
    t3.join()

通过设置 threading.Thread() 函数的 name 参数来设置线程名称，通过 threading.currentThread().getName() 来获取当前线程名称；线程的默认名称会以 Thread-i 格式来定义

自定义一个线程对象

示例代码如下所示

import threading
import time

exitFlag = 0


class myThread(threading.Thread):
    def __init__(self, threadID, name, counter):
        threading.Thread.__init__(self)
        self.threadID = threadID
        self.name = name
        self.counter = counter

    def run(self):
        print("Starting:0".format(self.name))
        print_time(self.name, self.counter, 5)
        print("Exiting:0".format(self.name))


def print_time(threadName, delay, counter):
    while counter:
        if exitFlag:
            thread.exit()
        time.sleep(delay)
        print("0 1".format(threadName, time.ctime(time.time())))
        counter -= 1


t1 = myThread(1, "Thread-1", 1)
t2 = myThread(2, "Thread-2", 1)

t1.start()
t2.start()

t1.join()
t2.join()

print("Exiting Main Thread.")

如果想自定义一个线程对象，首先就是要定义一个继承 threading.Thread 类的子类，实现构造函数， 并重写 run() 方法即可。

线程同步

Lock

示例代码如下所示

import threading

shared_resource_with_lock = 0
shared_resource_with_no_lock = 0
COUNT = 100000
shared_resource_lock = threading.Lock()


def increment_with_lock():
    global shared_resource_with_lock
    for i in range(COUNT):
        shared_resource_lock.acquire()
        shared_resource_with_lock += 1
        shared_resource_lock.release()


def decrement_with_lock():
    global shared_resource_with_lock
    for i in range(COUNT):
        shared_resource_lock.acquire()
        shared_resource_with_lock -= 1
        shared_resource_lock.release()


def increment_without_lock():
    global shared_resource_with_no_lock
    for i in range(COUNT):
        shared_resource_with_no_lock += 1


def decrement_wthout_lock():
    global shared_resource_with_no_lock
    for i in range(COUNT):
        shared_resource_with_no_lock -= 1


if __name__ == "__main__":
    t1 = threading.Thread(target=increment_with_lock)
    t2 = threading.Thread(target=decrement_with_lock)
    t3 = threading.Thread(target=increment_without_lock)
    t4 = threading.Thread(target=decrement_wthout_lock)
    t1.start()
    t2.start()
    t3.start()
    t4.start()
    t1.join()
    t2.join()
    t3.join()
    t4.join()
    print("the value of shared variable with lock management is :0".format(
        shared_resource_with_lock))
    print("the value of shared variable with race condition is :0".format(
        shared_resource_with_no_lock))

通过 threading.Lock() 方法我们可以拿到线程锁，一般有两种操作方式：acquire() 和 release() 在两者之间是加锁状态，如果释放失败的话会显示 RuntimError() 的异常。

RLock

RLock 也叫递归锁，和 Lock 的区别在于：谁拿到谁释放,是通过 threading.RLock() 来拿到的；

示例代码如下所示

import threading
import time


class Box(object):
    lock = threading.RLock()

    def __init__(self):
        self.total_items = 0

    def execute(self, n):
        Box.lock.acquire()
        self.total_items += n
        Box.lock.release()

    def add(self):
        Box.lock.acquire()
        self.execute(1)
        Box.lock.release()

    def remove(self):
        Box.lock.acquire()
        self.execute(-1)
        Box.lock.release()


def adder(box, items):
    while items > 0:
        print("adding 1 item in the box")
        box.add()
        time.sleep(1)
        items -= 1


def remover(box, items):
    while items > 0:
        print("removing 1 item in the box")
        box.remove()
        time.sleep(1)
        items -= 1


if __name__ == "__main__":
    items = 5
    print("putting 0 items in the box".format(items))
    box = Box()
    t1 = threading.Thread(target=adder, args=(box, items))
    t2 = threading.Thread(target=remover, args=(box, items))

    t1.start()
    t2.start()

    t1.join()
    t2.join()
    print("0 items still remain in the box".format(box.total_items))

信号量

示例代码如下所示

import threading
import time
import random

semaphore = threading.Semaphore(0)


def consumer():
    print("Consumer is waiting.")
    semaphore.acquire()
    print("Consumer notify:consumed item numbers 0".format(item))


def producer():
    global item
    time.sleep(10)
    item = random.randint(0, 10000)
    print("producer notify:produced item number 0".format(item))
    semaphore.release()


if __name__ == "__main__":
    for i in range(0, 5):
        t1 = threading.Thread(target=producer)
        t2 = threading.Thread(target=consumer)
        t1.start()
        t2.start()
        t1.join()
        t2.join()

    print("program terminated.")

信号量初始化为 0 ，然后在两个并行线程中，通过调用 semaphore.acquire() 函数会阻塞消费者线程，直到 semaphore.release() 在生产者中被调用，这里模拟了生产者-消费者 模式来进行了测试；如果信号量的计数器到了0，就会阻塞 acquire() 方法，直到得到另一个线程的通知。如果信号量的计数器大于0，就会对这个值-1然后分配资源。

使用条件进行线程同步

解释条件机制最好的例子还是生产者-消费者问题。在本例中，只要缓存不满，生产者一直向缓存生产；只要缓存不空，消费者一直从缓存取出（之后销毁）。当缓冲队列不为空的时候，生产者将通知消费者；当缓冲队列不满的时候，消费者将通知生产者。

示例代码如下所示

from threading import Thread, Condition
import time

items = []
condition = Condition()


class consumer(Thread):
    def __init__(self):
        Thread.__init__(self)

    def consume(self):
        global condition
        global items
        condition.acquire()
        if len(items) == 0:
            condition.wait()
            print("Consumer notify:no item to consum")
        items.pop()
        print("Consumer notify: consumed 1 item")
        print("Consumer notify: item to consume are:0".format(len(items)))

        condition.notify()
        condition.release()

    def run(self):
        for i in range(0, 20):
            time.sleep(2)
            self.consume()


class producer(Thread):
    def __init__(self):
        Thread.__init__(self)

    def produce(self):
        global condition
        global items
        condition.acquire()
        if len(items) == 10:
            condition.wait()
            print("Producer notify:items producted are:0".format(len(items)))
            print("Producer notify:stop the production!!")
        items.append(1)
        print("Producer notify:total items producted:0".format(len(items)))
        condition.notify()
        condition.release()

    def run(self):
        for i in range(0, 20):
            time.sleep(1)
            self.produce()


if __name__ == "__main__":
    producer = producer()
    consumer = consumer()
    producer.start()
    consumer.start()
    producer.join()
    consumer.join()

通过 condition.acquire() 来获取锁对象，condition.wait() 会使当前线程进入阻塞状态，直到收到 condition.notify() 信号，同时，调用信号的通知的对象也要及时调用 condition.release() 来释放资源；

使用事件进行线程同步

事件是线程之间用于通信的对。有的线程等待信号，有的线程发出信号。

示例代码如下所示

import time
from threading import Thread, Event
import random

items = []
event = Event()


class consumer(Thread):
    def __init__(self, items, event):
        Thread.__init__(self)
        self.items = items
        self.event = event

    def run(self):
        while True:
            time.sleep(2)
            self.event.wait()
            item = self.items.pop()
            print('Consumer notify:0 popped from list by 1'.format(
                item, self.name))


class producer(Thread):
    def __init__(self, integers, event):
        Thread.__init__(self)
        self.items = items
        self.event = event

    def run(self):
        global item
        for i in range(100):
            time.sleep(2)
            item = random.randint(0, 256)
            self.items.append(item)
            print('Producer notify: item  N° %d appended to list by %s' %
                  (item, self.name))
            print('Producer notify: event set by %s' % self.name)
            self.event.set()
            print('Produce notify: event cleared by %s ' % self.name)
            self.event.clear()


if __name__ == "__main__":
    t1 = producer(items, event)
    t2 = consumer(items, event)
    t1.start()
    t2.start()
    t1.join()
    t2.join()

使用 with 语法简化代码

import threading
import logging

logging.basicConfig(level=logging.DEBUG,
                    format='(%(threadName)-10s) %(message)s')


def threading_with(statement):
    with statement:
        logging.debug("%s acquired via with" % statement)


def Threading_not_with(statement):
    statement.acquire()
    try:
        logging.debug("%s acquired directly " % statement)
    finally:
        statement.release()


if __name__ == "__main__":
    lock = threading.Lock()
    rlock = threading.RLock()
    condition = threading.Condition()
    mutex = threading.Semaphore(1)
    threading_synchronization_list = [lock, rlock, condition, mutex]

    for statement in threading_synchronization_list:
        t1 = threading.Thread(target=threading_with, args=(statement,))
        t2 = threading.Thread(target=Threading_not_with, args=(statement,))
        t1.start()
        t2.start()
        t1.join()
        t2.join()

使用 queue 进行线程通信

Queue 常用的方法有以下四个：

• put()：往 queue 中添加一个元素
• get()：从 queue 中删除一个元素，并返回该元素
• task_done()：每次元素被处理的时候都需要调用这个方法
• join()：所有元素都被处理之前一直阻塞

from threading import Thread, Event
from queue import Queue
import time
import random


class producer(Thread):
    def __init__(self, queue):
        Thread.__init__(self)
        self.queue = queue

    def run(self):
        for i in range(10):
            item = random.randint(0, 256)
            self.queue.put(item)
            print("Producer notify: item item N° %d appended to queue by %s" %
                  (item, self.name))
            time.sleep(1)


class consumer(Thread):
    def __init__(self, queue):
        Thread.__init__(self)
        self.queue = queue

    def run(self):
        while True:
            item = self.queue.get()
            print('Consumer notify : %d popped from queue by %s' %
                  (item, self.name))
            self.queue.task_done()


if __name__ == "__main__":
    queue = Queue()
    t1 = producer(queue)
    t2 = consumer(queue)
    t3 = consumer(queue)
    t4 = consumer(queue)
    t1.start()
    t2.start()
    t3.start()
    t4.start()
    t1.join()
    t2.join()
    t3.join()
    t4.join()

基于进程的并行

multiprocessing 是 Python 标准库中的模块，实现了共享内存机制。

异步编程

使用 concurrent.futures 模块

该模块具有线程池和进程池，管理并行编程任务、处理非确定性的执行流程、进程/线程同步等功能；此模块由以下部分组成

• concurrent.futures.Executor: 这是一个虚拟基类，提供了异步执行的方法。
• submit(function, argument): 调度函数（可调用的对象）的执行，将 argument 作为参数传入。
• map(function, argument): 将 argument 作为参数执行函数，以 异步 的方式。
• shutdown(Wait=True): 发出让执行者释放所有资源的信号。
• concurrent.futures.Future: 其中包括函数的异步执行。Future对象是submit任务（即带有参数的functions）到executor的实例。

示例代码如下所示

import concurrent.futures
import time

number_list = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]


def evaluate_item(x):
    result_item = count(x)
    return result_item


def count(number):
    for i in range(0, 1000000):
        i = i + 1
    return i * number


if __name__ == "__main__":
    # 顺序执行
    start_time = time.time()
    for item in number_list:
        print(evaluate_item(item))
    print("Sequential execution in " + str(time.time() - start_time), "seconds")
    # 线程池执行
    start_time_1 = time.time()
    with concurrent.futures.ThreadPoolExecutor(max_workers=5) as executor:
        futures = [executor.submit(evaluate_item, item)
                   for item in number_list]
        for future in concurrent.futures.as_completed(futures):
            print(future.result())
    print("Thread pool execution in " +
          str(time.time() - start_time_1), "seconds")
    # 线程池执行
    start_time_2 = time.time()
    with concurrent.futures.ProcessPoolExecutor(max_workers=5) as executor:
        futures = [executor.submit(evaluate_item, item)
                   for item in number_list]
        for future in concurrent.futures.as_completed(futures):
            print(future.result())
    print("Process pool execution in " +
          str(time.time() - start_time_2), "seconds")

使用 Asyncio 管理事件循环

Python 的 Asyncio 模块提供了管理事件、协程、任务和线程的方法，以及编写并发代码的原语。此模块的主要组件和概念包括：

• 事件循环: 在Asyncio模块中，每一个进程都有一个事件循环。
• 协程: 这是子程序的泛化概念。协程可以在执行期间暂停，这样就可以等待外部的处理（例如IO）完成之后，从之前暂停的地方恢复执行。
• Futures: 定义了 Future 对象，和 concurrent.futures 模块一样，表示尚未完成的计算。
• Tasks: 这是Asyncio的子类，用于封装和管理并行模式下的协程。

Asyncio 提供了以下方法来管理事件循环：

• loop = get_event_loop(): 得到当前上下文的事件循环。
• loop.call_later(time_delay, callback, argument): 延后 time_delay 秒再执行 callback 方法。
• loop.call_soon(callback, argument): 尽可能快调用 callback, call_soon() 函数结束，主线程回到事件循环之后就会马上调用 callback 。
• loop.time(): 以float类型返回当前时间循环的内部时间。
• asyncio.set_event_loop(): 为当前上下文设置事件循环。
• asyncio.new_event_loop(): 根据此策略创建一个新的时间循环并返回。
• loop.run_forever(): 在调用 stop() 之前将一直运行。

示例代码如下所示

import asyncio
import datetime
import time


def fuction_1(end_time, loop):
    print("function_1 called")
    if(loop.time() + 1.0) < end_time:
        loop.call_later(1, fuction_2, end_time, loop)
    else:
        loop.stop()


def fuction_2(end_time, loop):
    print("function_2 called")
    if(loop.time() + 1.0) < end_time:
        loop.call_later(1, function_3, end_time, loop)
    else:
        loop.stop()


def function_3(end_time, loop):
    print("function_3 called")
    if(loop.time() + 1.0) < end_time:
        loop.call_later(1, fuction_1, end_time, loop)
    else:
        loop.stop()


def function_4(end_time, loop):
    print("function_4 called")
    if(loop.time() + 1.0) < end_time:
        loop.call_later(1, function_4, end_time, loop)
    else:
        loop.stop()


loop = asyncio.get_event_loop()

end_loop = loop.time() + 9.0
loop.call_soon(fuction_1, end_loop, loop)
loop.run_forever()
loop.close()

使用 Asyncio 管理协程

示例代码如下所示

import asyncio
import time
from random import randint


@asyncio.coroutine
def StartState():
    print("Start State called \n")
    input_val = randint(0, 1)
    time.sleep(1)
    if input_val == 0:
        result = yield from State2(input_val)
    else:
        result = yield from State1(input_val)
    print("Resume of the Transition:\nStart State calling" + result)


@asyncio.coroutine
def State1(transition_value):
    outputVal = str("State 1 with transition value=%s \n" % (transition_value))
    input_val = randint(0, 1)
    time.sleep(1)
    print("...Evaluating...")
    if input_val == 0:
        result = yield from State3(input_val)
    else:
        result = yield from State2(input_val)


@asyncio.coroutine
def State2(transition_value):
    outputVal = str("State 2 with transition value= %s \n" %
                    (transition_value))
    input_Val = randint(0, 1)
    time.sleep(1)
    print("...Evaluating...")
    if (input_Val == 0):
        result = yield from State1(input_Val)
    else:
        result = yield from State3(input_Val)
    result = "State 2 calling " + result
    return outputVal + str(result)


@asyncio.coroutine
def State3(transition_value):
    outputVal = str("State 3 with transition value = %s \n" %
                    (transition_value))
    input_val = randint(0, 1)
    time.sleep(1)
    print("...Evaluating...")
    if(input_val == 0):
        result = yield from State1(input_val)
    else:
        result = yield from State2(input_val)
    result = "State 3 calling " + result
    return outputVal + str(result)


@asyncio.coroutine
def EndState(transition_value):
    outputVal = str("End State With transition value = %s \n" %
                    (transition_value))
    print("...Stop Computation...")
    return outputVal


if __name__ == "__main__":
    print("Finites State Machine simulation with Asyncio Coroutine")
    loop = asyncio.get_event_loop()
    loop.run_until_complete(StartState())

使用 Asyncio 控制任务

示例代码如下所示

import asyncio


@asyncio.coroutine
def factorial(number):
    f = 1
    for i in range(2, number + 1):
        print("Asyncio.Task:Compute factorial(%s)" % (i))
        yield from asyncio.sleep(1)
        f *= i
    print("Asyncio.Task - factorial(%s) = %s" % (number, f))


@asyncio.coroutine
def fibonacci(number):
    a, b = 0, 1
    for i in range(number):
        print("Asyncio.Task:Complete fibonacci (%s)" % (i))
        yield from asyncio.sleep(1)
        a, b = b, a+b
    print("Asyncio.Task - fibonaci (%s)= %s" % (number, a))


@asyncio.coroutine
def binomialCoeff(n, k):
    result = 1
    for i in range(1, k+1):
        result = result * (n-i+1) / i
        print("Asyncio.Task:Compute binomialCoeff (%s)" % (i))
        yield from asyncio.sleep(1)
    print("Asyncio.Task - binomialCoeff (%s,%s) = %s" % (n, k, result))


if __name__ == "__main__":
    tasks = [asyncio.Task(factorial(10)), asyncio.Task(
        fibonacci(10)), asyncio.Task(binomialCoeff(20, 10))]
    loop = asyncio.get_event_loop()
    loop.run_until_complete(asyncio.wait(tasks))
    loop.close()

使用Asyncio和Futures

示例代码如下所示

import asyncio
import sys


@asyncio.coroutine
def first_coroutine(future, N):
    count = 0
    for i in range(1, N + 1):
        count = count + i
    yield from asyncio.sleep(4)
    future.set_result(
        "first coroutine (sum of N integers) result = " + str(count))


@asyncio.coroutine
def second_coroutine(future, N):
    count = 1
    for i in range(2, N + 1):
        count *= i
    yield from asyncio.sleep(3)
    future.set_result("second coroutine (factorial) result = " + str(count))


def got_result(future):
    print(future.result())


if __name__ == "__main__":
    N1 = 1
    N2 = 1
    loop = asyncio.get_event_loop()
    future1 = asyncio.Future()
    future2 = asyncio.Future()
    tasks = [
        first_coroutine(future1, N1),
        second_coroutine(future2, N2)
    ]
    future1.add_done_callback(got_result)
    future2.add_done_callback(got_result)
    loop.run_until_complete(asyncio.wait(tasks))
    loop.close()

分布式编程

略

GPU 编程

略

相关参考

• Python并行编程 中文版