关于python:限制/限制GRequests中HTTP请求的速率

Limiting/throttling the rate of HTTP requests in GRequests

我正在用python 2.7.3编写一个小脚本,其中包含grequests和lxml,它允许我从各种网站收集一些可收集的卡片价格,并对它们进行比较。问题是其中一个网站限制了请求的数量,如果超过了这个数量,就会返回HTTP错误429。

是否有一种方法可以添加限制grequest中的请求数,以使我不超过每秒指定的请求数?另外-如果发生HTTP 429,如何在一段时间后使grequestes重试?

旁注——他们的限额低得离谱。大约每15秒有8个请求。我多次使用浏览器破坏它,只是刷新等待价格变化的页面。


我要自己回答我自己的问题,因为我必须自己解决这个问题,而且这方面的信息似乎很少。

想法如下。与grequest一起使用的每个请求对象在创建时都可以将会话对象作为参数。另一方面,会话对象可以安装HTTP适配器,在发出请求时使用这些适配器。通过创建我们自己的适配器,我们可以截获请求,并以最适合我们应用程序的方式对请求进行速率限制。在我的例子中,我最后得到了下面的代码。

用于限制的对象:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
DEFAULT_BURST_WINDOW = datetime.timedelta(seconds=5)
DEFAULT_WAIT_WINDOW = datetime.timedelta(seconds=15)


class BurstThrottle(object):
    max_hits = None
    hits = None
    burst_window = None
    total_window = None
    timestamp = None

    def __init__(self, max_hits, burst_window, wait_window):
        self.max_hits = max_hits
        self.hits = 0
        self.burst_window = burst_window
        self.total_window = burst_window + wait_window
        self.timestamp = datetime.datetime.min

    def throttle(self):
        now = datetime.datetime.utcnow()
        if now < self.timestamp + self.total_window:
            if (now < self.timestamp + self.burst_window) and (self.hits < self.max_hits):
                self.hits += 1
                return datetime.timedelta(0)
            else:
                return self.timestamp + self.total_window - now
        else:
            self.timestamp = now
            self.hits = 1
            return datetime.timedelta(0)

HTTP适配器:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
class MyHttpAdapter(requests.adapters.HTTPAdapter):
    throttle = None

    def __init__(self, pool_connections=requests.adapters.DEFAULT_POOLSIZE,
                 pool_maxsize=requests.adapters.DEFAULT_POOLSIZE, max_retries=requests.adapters.DEFAULT_RETRIES,
                 pool_block=requests.adapters.DEFAULT_POOLBLOCK, burst_window=DEFAULT_BURST_WINDOW,
                 wait_window=DEFAULT_WAIT_WINDOW):
        self.throttle = BurstThrottle(pool_maxsize, burst_window, wait_window)
        super(MyHttpAdapter, self).__init__(pool_connections=pool_connections, pool_maxsize=pool_maxsize,
                                            max_retries=max_retries, pool_block=pool_block)

    def send(self, request, stream=False, timeout=None, verify=True, cert=None, proxies=None):
        request_successful = False
        response = None
        while not request_successful:
            wait_time = self.throttle.throttle()
            while wait_time > datetime.timedelta(0):
                gevent.sleep(wait_time.total_seconds(), ref=True)
                wait_time = self.throttle.throttle()

            response = super(MyHttpAdapter, self).send(request, stream=stream, timeout=timeout,
                                                       verify=verify, cert=cert, proxies=proxies)

            if response.status_code != 429:
                request_successful = True

        return response

设置:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
requests_adapter = adapter.MyHttpAdapter(
    pool_connections=__CONCURRENT_LIMIT__,
    pool_maxsize=__CONCURRENT_LIMIT__,
    max_retries=0,
    pool_block=False,
    burst_window=datetime.timedelta(seconds=5),
    wait_window=datetime.timedelta(seconds=20))

requests_session = requests.session()
requests_session.mount('http://', requests_adapter)
requests_session.mount('https://', requests_adapter)

unsent_requests = (grequests.get(url,
                                 hooks={'response': handle_response},
                                 session=requests_session) for url in urls)
grequests.map(unsent_requests, size=__CONCURRENT_LIMIT__)

请看一下自动请求限制:https://pypi.python.org/pypi/requeststhrottler/0.2.2

您可以在每个请求之间设置固定的延迟量,也可以在固定的秒数内设置发送的请求数(基本上相同):

1
2
3
4
5
6
7
import requests
from requests_throttler import BaseThrottler

request = requests.Request(method='GET', url='http://www.google.com')
reqs = [request for i in range(0, 5)]  # An example list of requests
with BaseThrottler(name='base-throttler', delay=1.5) as bt:
    throttled_requests = bt.multi_submit(reqs)

其中函数multi_submit返回ThrottledRequest的列表(见末尾的doc:link)。

然后您可以访问响应:

1
2
for tr in throttled_requests:
    print tr.response

或者,您可以通过指定在固定时间内发送的一个或多个请求(例如,每60秒发送15个请求)来实现这一点:

1
2
3
4
5
6
7
import requests
from requests_throttler import BaseThrottler

request = requests.Request(method='GET', url='http://www.google.com')
reqs = [request for i in range(0, 5)]  # An example list of requests
with BaseThrottler(name='base-throttler', reqs_over_time=(15, 60)) as bt:
    throttled_requests = bt.multi_submit(reqs)

两种解决方案都可以在不使用with语句的情况下实现:

1
2
3
4
5
6
7
8
9
import requests
from requests_throttler import BaseThrottler

request = requests.Request(method='GET', url='http://www.google.com')
reqs = [request for i in range(0, 5)]  # An example list of requests
bt = BaseThrottler(name='base-throttler', delay=1.5)
bt.start()
throttled_requests = bt.multi_submit(reqs)
bt.shutdown()

有关详细信息,请访问:http://pythonhosted.org/requeststhrottler/index.html


我也有类似的问题。这是我的解决方案。在你的情况下,我会:

1
2
3
4
5
def worker():
    with rate_limit('slow.domain.com', 2):
        response = requests.get('https://slow.domain.com/path')
        text = response.text
    # Use `text`

假设你有多个域名,我会设置一个字典映射(domain, delay),这样你就不会达到你的速率限制。

此代码假定您将使用gevent和monkey补丁。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
from contextlib import contextmanager
from gevent.event import Event
from gevent.queue import Queue
from time import time


def rate_limit(resource, delay, _queues={}):
   """Delay use of `resource` until after `delay` seconds have passed.

    Example usage:

    def worker():
        with rate_limit('foo.bar.com', 1):
            response = requests.get('https://foo.bar.com/path')
            text = response.text
        # use `text`

    This will serialize and delay requests from multiple workers for resource
    'foo.bar.com' by 1 second.

   """


    if resource not in _queues:
        queue = Queue()
        gevent.spawn(_watch, queue)
        _queues[resource] = queue

    return _resource_manager(_queues[resource], delay)


def _watch(queue):
   "Watch `queue` and wake event listeners after delay."

    last = 0

    while True:
        event, delay = queue.get()

        now = time()

        if (now - last) < delay:
            gevent.sleep(delay - (now - last))

        event.set()   # Wake worker but keep control.
        event.clear()
        event.wait()  # Yield control until woken.

        last = time()


@contextmanager
def _resource_manager(queue, delay):
   "`with` statement support for `rate_limit`."

    event = Event()
    queue.put((event, delay))

    event.wait() # Wait for queue watcher to wake us.

    yield

    event.set()  # Wake queue watcher.

看起来没有任何简单的机制来处理请求或grequests代码中的这个内置项。唯一的钩子似乎是周围的反应。

这里有一个超级黑客的工作,至少证明了这是可能的-我修改了grequest,以保留发出请求的时间列表,并休眠异步请求的创建,直到每秒请求数低于最大值。

1
2
3
4
5
6
7
8
9
10
11
class AsyncRequest(object):
    def __init__(self, method, url, **kwargs):
        print self,'init'
        waiting=True
        while waiting:
            if len([x for x in q if x > time.time()-15]) < 8:
                q.append(time.time())
                waiting=False
            else:
                print self,'snoozing'
                gevent.sleep(1)

您可以使用grequests.imap()以交互方式观看

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
import time
import rg

urls = [
        'http://www.heroku.com',
        'http://python-tablib.org',
        'http://httpbin.org',
        'http://python-requests.org',
        'http://kennethreitz.com',
        'http://www.cnn.com',
]

def print_url(r, *args, **kwargs):
        print(r.url),time.time()

hook_dict=dict(response=print_url)
rs = (rg.get(u, hooks=hook_dict) for u in urls)
for r in rg.imap(rs):
        print r

我希望有一个更优雅的解决方案,但到目前为止我找不到。在会话和适配器中四处查看。也许游泳池管理员可以增加?

此外,我不会将此代码投入生产中,'q'列表永远不会被修剪,最终会变得相当大。另外,我不知道它是否真的像广告上说的那样工作。当我查看控制台输出时,它看起来就是这样。

呃。只要看看这段代码,我就知道是凌晨3点。该睡觉了。