376 lines
13 KiB
Markdown
376 lines
13 KiB
Markdown
|
## 并发下载
|
|||
|
|
|||
|
### 多线程和多进程回顾
|
|||
|
|
|||
|
在前面的[《进程和线程》](../Day01-15/Day13/进程和线程.md)一文中,我们已经对在Python中使用多进程和多线程实现并发编程进行了简明的讲解,在此我们补充几个知识点。
|
|||
|
|
|||
|
#### threading.local类
|
|||
|
|
|||
|
使用线程时最不愿意遇到的情况就是多个线程竞争资源,在这种情况下为了保证资源状态的正确性,我们可能需要对资源进行加锁保护的处理,这一方面会导致程序失去并发性,另外如果多个线程竞争多个资源时,还有可能因为加锁方式的不当导致[死锁](https://zh.wikipedia.org/wiki/%E6%AD%BB%E9%94%81)。要解决多个线程竞争资源的问题,其中一个方案就是让每个线程都持有资源的副本(拷贝),这样每个线程可以操作自己所持有的资源,从而规避对资源的竞争。
|
|||
|
|
|||
|
要实现将资源和持有资源的线程进行绑定的操作,最简单的做法就是使用threading模块的local类,在网络爬虫开发中,就可以使用local类为每个线程绑定一个MySQL数据库连接或Redis客户端对象,这样通过线程可以直接获得这些资源,既解决了资源竞争的问题,又避免了在函数和方法调用时传递这些资源。具体的请参考本章多线程爬取“手机搜狐网”(Redis版)的实例代码。
|
|||
|
|
|||
|
#### concurrent.futures模块
|
|||
|
|
|||
|
Python3.2带来了`concurrent.futures` 模块,这个模块包含了线程池和进程池、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。关于这部分的内容推荐大家阅读[《Python并行编程》](http://python-parallel-programmning-cookbook.readthedocs.io/zh_CN/latest/index.html)。
|
|||
|
|
|||
|
#### 分布式进程
|
|||
|
|
|||
|
使用多进程的时候,可以将进程部署在多个主机节点上,Python的`multiprocessing`模块不但支持多进程,其中`managers`子模块还支持把多进程部署到多个节点上。当然,要部署分布式进程,首先需要一个服务进程作为调度者,进程之间通过网络进行通信来实现对进程的控制和调度,由于`managers`模块已经对这些做出了很好的封装,因此在无需了解网络通信细节的前提下,就可以编写分布式多进程应用。具体的请参照本章分布式多进程爬取“手机搜狐网”的实例代码。
|
|||
|
|
|||
|
### 协程和异步I/O
|
|||
|
|
|||
|
#### 协程的概念
|
|||
|
|
|||
|
协程(coroutine)通常又称之为微线程或纤程,它是相互协作的一组子程序(函数)。所谓相互协作指的是在执行函数A时,可以随时中断去执行函数B,然后又中断继续执行函数A。注意,这一过程并不是函数调用(因为没有调用语句),整个过程看似像多线程,然而协程只有一个线程执行。协程通过`yield`关键字和 `send()`操作来转移执行权,协程之间不是调用者与被调用者的关系。
|
|||
|
|
|||
|
协程的优势在于以下两点:
|
|||
|
|
|||
|
1. 执行效率极高,因为子程序(函数)切换不是线程切换,由程序自身控制,没有切换线程的开销。
|
|||
|
2. 不需要多线程的锁机制,因为只有一个线程,也不存在竞争资源的问题,当然也就不需要对资源加锁保护,因此执行效率高很多。
|
|||
|
|
|||
|
> 说明:协程适合处理的是I/O密集型任务,处理CPU密集型任务并不是它的长处,如果要提升CPU的利用率可以考虑“多进程+协程”的模式。
|
|||
|
|
|||
|
#### 历史回顾
|
|||
|
|
|||
|
1. Python 2.2:第一次提出了生成器(最初称之为迭代器)的概念(PEP 255)。
|
|||
|
2. Python 2.5:引入了将对象发送回暂停了的生成器这一特性即生成器的`send()`方法(PEP 342)。
|
|||
|
3. Python 3.3:添加了`yield from`特性,允许从迭代器中返回任何值(注意生成器本身也是迭代器),这样我们就可以串联生成器并且重构出更好的生成器。
|
|||
|
4. Python 3.4:引入`asyncio.coroutine`装饰器用来标记作为协程的函数,协程函数和`asyncio`及其事件循环一起使用,来实现异步I/O操作。
|
|||
|
5. Python 3.5:引入了`async`和`await`,可以使用`async def`来定义一个协程函数,这个函数中不能包含任何形式的`yield`语句,但是可以使用`return`或`await`从协程中返回值。
|
|||
|
|
|||
|
#### 示例代码
|
|||
|
|
|||
|
1. 生成器 - 数据的生产者。
|
|||
|
|
|||
|
```Python
|
|||
|
from time import sleep
|
|||
|
|
|||
|
|
|||
|
# 倒计数生成器
|
|||
|
def countdown(n):
|
|||
|
while n > 0:
|
|||
|
yield n
|
|||
|
n -= 1
|
|||
|
|
|||
|
|
|||
|
def main():
|
|||
|
for num in countdown(5):
|
|||
|
print(f'Countdown: {num}')
|
|||
|
sleep(1)
|
|||
|
print('Countdown Over!')
|
|||
|
|
|||
|
|
|||
|
if __name__ == '__main__':
|
|||
|
main()
|
|||
|
|
|||
|
```
|
|||
|
|
|||
|
生成器还可以叠加来组成生成器管道,代码如下所示。
|
|||
|
|
|||
|
```Python
|
|||
|
# Fibonacci数生成器
|
|||
|
def fib():
|
|||
|
a, b = 0, 1
|
|||
|
while True:
|
|||
|
a, b = b, a + b
|
|||
|
yield a
|
|||
|
|
|||
|
|
|||
|
# 偶数生成器
|
|||
|
def even(gen):
|
|||
|
for val in gen:
|
|||
|
if val % 2 == 0:
|
|||
|
yield val
|
|||
|
|
|||
|
|
|||
|
def main():
|
|||
|
gen = even(fib())
|
|||
|
for _ in range(10):
|
|||
|
print(next(gen))
|
|||
|
|
|||
|
|
|||
|
if __name__ == '__main__':
|
|||
|
main()
|
|||
|
```
|
|||
|
|
|||
|
2. 协程 - 数据的消费者。
|
|||
|
|
|||
|
```Python
|
|||
|
from time import sleep
|
|||
|
|
|||
|
|
|||
|
# 生成器 - 数据生产者
|
|||
|
def countdown_gen(n, consumer):
|
|||
|
consumer.send(None)
|
|||
|
while n > 0:
|
|||
|
consumer.send(n)
|
|||
|
n -= 1
|
|||
|
consumer.send(None)
|
|||
|
|
|||
|
|
|||
|
# 协程 - 数据消费者
|
|||
|
def countdown_con():
|
|||
|
while True:
|
|||
|
n = yield
|
|||
|
if n:
|
|||
|
print(f'Countdown {n}')
|
|||
|
sleep(1)
|
|||
|
else:
|
|||
|
print('Countdown Over!')
|
|||
|
|
|||
|
|
|||
|
def main():
|
|||
|
countdown_gen(5, countdown_con())
|
|||
|
|
|||
|
|
|||
|
if __name__ == '__main__':
|
|||
|
main()
|
|||
|
```
|
|||
|
|
|||
|
> 说明:上面代码中countdown_gen函数中的第1行consumer.send(None)是为了激活生成器,通俗的说就是让生成器执行到有yield关键字的地方挂起,当然也可以通过next(consumer)来达到同样的效果。如果不愿意每次都用这样的代码来“预激”生成器,可以写一个包装器来完成该操作,代码如下所示。
|
|||
|
|
|||
|
```Python
|
|||
|
from functools import wraps
|
|||
|
|
|||
|
|
|||
|
def coroutine(fn):
|
|||
|
|
|||
|
@wraps(fn)
|
|||
|
def wrapper(*args, **kwargs):
|
|||
|
gen = fn(*args, **kwargs)
|
|||
|
next(gen)
|
|||
|
return gen
|
|||
|
|
|||
|
return wrapper
|
|||
|
```
|
|||
|
|
|||
|
这样就可以使用`@coroutine`装饰器对协程进行预激操作,不需要再写重复代码来激活协程。
|
|||
|
|
|||
|
3. 异步I/O - 非阻塞式I/O操作。
|
|||
|
|
|||
|
```Python
|
|||
|
import asyncio
|
|||
|
|
|||
|
|
|||
|
@asyncio.coroutine
|
|||
|
def countdown(name, n):
|
|||
|
while n > 0:
|
|||
|
print(f'Countdown[{name}]: {n}')
|
|||
|
yield from asyncio.sleep(1)
|
|||
|
n -= 1
|
|||
|
|
|||
|
|
|||
|
def main():
|
|||
|
loop = asyncio.get_event_loop()
|
|||
|
tasks = [
|
|||
|
countdown("A", 10), countdown("B", 5),
|
|||
|
]
|
|||
|
loop.run_until_complete(asyncio.wait(tasks))
|
|||
|
loop.close()
|
|||
|
|
|||
|
|
|||
|
if __name__ == '__main__':
|
|||
|
main()
|
|||
|
```
|
|||
|
|
|||
|
4. `async`和`await`。
|
|||
|
|
|||
|
```Python
|
|||
|
import asyncio
|
|||
|
import aiohttp
|
|||
|
|
|||
|
|
|||
|
async def download(url):
|
|||
|
print('Fetch:', url)
|
|||
|
async with aiohttp.ClientSession() as session:
|
|||
|
async with session.get(url) as resp:
|
|||
|
print(url, '--->', resp.status)
|
|||
|
print(url, '--->', resp.cookies)
|
|||
|
print('\n\n', await resp.text())
|
|||
|
|
|||
|
|
|||
|
def main():
|
|||
|
loop = asyncio.get_event_loop()
|
|||
|
urls = [
|
|||
|
'https://www.baidu.com',
|
|||
|
'http://www.sohu.com/',
|
|||
|
'http://www.sina.com.cn/',
|
|||
|
'https://www.taobao.com/',
|
|||
|
'https://www.jd.com/'
|
|||
|
]
|
|||
|
tasks = [download(url) for url in urls]
|
|||
|
loop.run_until_complete(asyncio.wait(tasks))
|
|||
|
loop.close()
|
|||
|
|
|||
|
|
|||
|
if __name__ == '__main__':
|
|||
|
main()
|
|||
|
```
|
|||
|
|
|||
|
上面的代码使用了[AIOHTTP](https://github.com/aio-libs/aiohttp)这个非常著名的第三方库,它实现了HTTP客户端和HTTP服务器的功能,对异步操作提供了非常好的支持,有兴趣可以阅读它的[官方文档](https://aiohttp.readthedocs.io/en/stable/)。
|
|||
|
|
|||
|
### 实例 - 多线程爬取“手机搜狐网”所有页面
|
|||
|
|
|||
|
下面我们把之间讲的所有知识结合起来,用面向对象的方式实现一个爬取“手机搜狐网”的多线程爬虫。
|
|||
|
|
|||
|
```Python
|
|||
|
import pickle
|
|||
|
import zlib
|
|||
|
from enum import Enum, unique
|
|||
|
from hashlib import sha1
|
|||
|
from random import random
|
|||
|
from threading import Thread, current_thread, local
|
|||
|
from time import sleep
|
|||
|
from urllib.parse import urlparse
|
|||
|
|
|||
|
import pymongo
|
|||
|
import redis
|
|||
|
import requests
|
|||
|
from bs4 import BeautifulSoup
|
|||
|
from bson import Binary
|
|||
|
|
|||
|
|
|||
|
@unique
|
|||
|
class SpiderStatus(Enum):
|
|||
|
IDLE = 0
|
|||
|
WORKING = 1
|
|||
|
|
|||
|
|
|||
|
def decode_page(page_bytes, charsets=('utf-8',)):
|
|||
|
page_html = None
|
|||
|
for charset in charsets:
|
|||
|
try:
|
|||
|
page_html = page_bytes.decode(charset)
|
|||
|
break
|
|||
|
except UnicodeDecodeError:
|
|||
|
pass
|
|||
|
return page_html
|
|||
|
|
|||
|
|
|||
|
class Retry(object):
|
|||
|
|
|||
|
def __init__(self, *, retry_times=3,
|
|||
|
wait_secs=5, errors=(Exception, )):
|
|||
|
self.retry_times = retry_times
|
|||
|
self.wait_secs = wait_secs
|
|||
|
self.errors = errors
|
|||
|
|
|||
|
def __call__(self, fn):
|
|||
|
|
|||
|
def wrapper(*args, **kwargs):
|
|||
|
for _ in range(self.retry_times):
|
|||
|
try:
|
|||
|
return fn(*args, **kwargs)
|
|||
|
except self.errors as e:
|
|||
|
print(e)
|
|||
|
sleep((random() + 1) * self.wait_secs)
|
|||
|
return None
|
|||
|
|
|||
|
return wrapper
|
|||
|
|
|||
|
|
|||
|
class Spider(object):
|
|||
|
|
|||
|
def __init__(self):
|
|||
|
self.status = SpiderStatus.IDLE
|
|||
|
|
|||
|
@Retry()
|
|||
|
def fetch(self, current_url, *, charsets=('utf-8', ),
|
|||
|
user_agent=None, proxies=None):
|
|||
|
thread_name = current_thread().name
|
|||
|
print(f'[{thread_name}]: {current_url}')
|
|||
|
headers = {'user-agent': user_agent} if user_agent else {}
|
|||
|
resp = requests.get(current_url,
|
|||
|
headers=headers, proxies=proxies)
|
|||
|
return decode_page(resp.content, charsets) \
|
|||
|
if resp.status_code == 200 else None
|
|||
|
|
|||
|
def parse(self, html_page, *, domain='m.sohu.com'):
|
|||
|
soup = BeautifulSoup(html_page, 'lxml')
|
|||
|
for a_tag in soup.body.select('a[href]'):
|
|||
|
parser = urlparse(a_tag.attrs['href'])
|
|||
|
scheme = parser.scheme or 'http'
|
|||
|
netloc = parser.netloc or domain
|
|||
|
if scheme != 'javascript' and netloc == domain:
|
|||
|
path = parser.path
|
|||
|
query = '?' + parser.query if parser.query else ''
|
|||
|
full_url = f'{scheme}://{netloc}{path}{query}'
|
|||
|
redis_client = thread_local.redis_client
|
|||
|
if not redis_client.sismember('visited_urls', full_url):
|
|||
|
redis_client.rpush('m_sohu_task', full_url)
|
|||
|
|
|||
|
def extract(self, html_page):
|
|||
|
pass
|
|||
|
|
|||
|
def store(self, data_dict):
|
|||
|
# redis_client = thread_local.redis_client
|
|||
|
# mongo_db = thread_local.mongo_db
|
|||
|
pass
|
|||
|
|
|||
|
|
|||
|
class SpiderThread(Thread):
|
|||
|
|
|||
|
def __init__(self, name, spider):
|
|||
|
super().__init__(name=name, daemon=True)
|
|||
|
self.spider = spider
|
|||
|
|
|||
|
def run(self):
|
|||
|
redis_client = redis.Redis(host='1.2.3.4', port=6379, password='1qaz2wsx')
|
|||
|
mongo_client = pymongo.MongoClient(host='1.2.3.4', port=27017)
|
|||
|
thread_local.redis_client = redis_client
|
|||
|
thread_local.mongo_db = mongo_client.msohu
|
|||
|
while True:
|
|||
|
current_url = redis_client.lpop('m_sohu_task')
|
|||
|
while not current_url:
|
|||
|
current_url = redis_client.lpop('m_sohu_task')
|
|||
|
self.spider.status = SpiderStatus.WORKING
|
|||
|
current_url = current_url.decode('utf-8')
|
|||
|
if not redis_client.sismember('visited_urls', current_url):
|
|||
|
redis_client.sadd('visited_urls', current_url)
|
|||
|
html_page = self.spider.fetch(current_url)
|
|||
|
if html_page not in [None, '']:
|
|||
|
hasher = hasher_proto.copy()
|
|||
|
hasher.update(current_url.encode('utf-8'))
|
|||
|
doc_id = hasher.hexdigest()
|
|||
|
sohu_data_coll = mongo_client.msohu.webpages
|
|||
|
if not sohu_data_coll.find_one({'_id': doc_id}):
|
|||
|
sohu_data_coll.insert_one({
|
|||
|
'_id': doc_id,
|
|||
|
'url': current_url,
|
|||
|
'page': Binary(zlib.compress(pickle.dumps(html_page)))
|
|||
|
})
|
|||
|
self.spider.parse(html_page)
|
|||
|
self.spider.status = SpiderStatus.IDLE
|
|||
|
|
|||
|
|
|||
|
def is_any_alive(spider_threads):
|
|||
|
return any([spider_thread.spider.status == SpiderStatus.WORKING
|
|||
|
for spider_thread in spider_threads])
|
|||
|
|
|||
|
|
|||
|
thread_local = local()
|
|||
|
hasher_proto = sha1()
|
|||
|
|
|||
|
|
|||
|
def main():
|
|||
|
redis_client = redis.Redis(host='1.2.3.4', port=6379, password='1qaz2wsx')
|
|||
|
if not redis_client.exists('m_sohu_task'):
|
|||
|
redis_client.rpush('m_sohu_task', 'http://m.sohu.com/')
|
|||
|
|
|||
|
spider_threads = [SpiderThread('thread-%d' % i, Spider())
|
|||
|
for i in range(10)]
|
|||
|
for spider_thread in spider_threads:
|
|||
|
spider_thread.start()
|
|||
|
|
|||
|
while redis_client.exists('m_sohu_task') or is_any_alive(spider_threads):
|
|||
|
sleep(5)
|
|||
|
|
|||
|
print('Over!')
|
|||
|
|
|||
|
|
|||
|
if __name__ == '__main__':
|
|||
|
main()
|
|||
|
```
|
|||
|
|