gevent/grequests 通过 HTTPS 的奇怪阻塞行为

Question

以下代码每 200 毫秒发送一个请求，并且应该在响应到来时异步处理它们。

通过 HTTP 它按预期工作 - 每 200 毫秒发送一个请求，并且每当响应到达时独立调用响应回调。然而，通过 HTTPS，请求会在响应到达时显着延迟（即使我的响应处理程序不起作用）。似乎每个请求都调用了两次响应回调，一次是零长度响应（编辑：这是因为重定向，似乎与阻塞问题无关，感谢 Padraic）。

什么可能导致这种 HTTPS 阻塞行为？（www.bbc.co.uk 只是一个在地理上离我很远的例子，但它发生在我测试过的所有服务器上）。

grequests_test.py

import time
import sys
import grequests
import gevent

def cb(res, **kwargs):
    print("**** Response", time.time(), len(res.text))

for i in range(10):
    unsent = grequests.get(sys.argv[1], hooks={'response': cb})
    print("Request", time.time())
    grequests.send(unsent, grequests.Pool(1))
    gevent.sleep(0.2)
gevent.sleep(5)

$ ipython2 grequests_test.py 'http://www.bbc.co.uk'（预期结果）

('Request', 1459050191.499266)
('Request', 1459050191.701466)
('Request', 1459050191.903223)
('Request', 1459050192.10403)
('Request', 1459050192.305626)
('**** Response', 1459050192.099185, 179643)
('Request', 1459050192.506476)
('**** Response', 1459050192.307869, 179643)
('Request', 1459050192.707745)
('**** Response', 1459050192.484711, 179643)
('Request', 1459050192.909376)
('**** Response', 1459050192.696583, 179643)
('Request', 1459050193.110528)
('**** Response', 1459050192.870476, 179643)
('Request', 1459050193.311601)
('**** Response', 1459050193.071679, 179639)
('**** Response', 1459050193.313615, 179680)
('**** Response', 1459050193.4959, 179643)
('**** Response', 1459050193.687054, 179680)
('**** Response', 1459050193.902827, 179639)

ipython2 grequests_test.py 'https://www.bbc.co.uk'（请求发送晚）

('Request', 1459050203.24336)
('Request', 1459050203.44473)
('**** Response', 1459050204.423302, 0)
('Request', 1459050204.424748) <------------- THIS REQUEST TIME IS LATE
('**** Response', 1459050205.294426, 0)
('Request', 1459050205.296722)
('Request', 1459050205.497924)
('**** Response', 1459050206.456572, 0)
('Request', 1459050206.457875)
('**** Response', 1459050207.363188, 0)
('**** Response', 1459050208.247189, 0)
('Request', 1459050208.249579)
('**** Response', 1459050208.250645, 179643)
('**** Response', 1459050208.253638, 179643)
('Request', 1459050208.451083)
('**** Response', 1459050209.426556, 0)
('Request', 1459050209.428032)
('**** Response', 1459050209.428929, 179643)
('**** Response', 1459050210.331425, 0)
('**** Response', 1459050211.247793, 0)
('Request', 1459050211.251574)
('**** Response', 1459050211.252321, 179643)
('**** Response', 1459050211.25519, 179680)
('**** Response', 1459050212.397186, 0)
('**** Response', 1459050213.299109, 0)
('**** Response', 1459050213.588854, 179588)
('**** Response', 1459050213.590434, 179643)
('**** Response', 1459050213.593731, 179643)
('**** Response', 1459050213.90507, 179643)
('**** Response', 1459050213.909386, 179643)

请注意，第一个响应似乎在下一个请求应该已发送但未发送后很久才到达。为什么在第一个响应到达之前没有睡眠 return 和下一个请求发送？

Answer 1

额外的响应和 0 长度的响应很容易解释，如果你添加一个 print(res.status_code) 你会看到很多 301，在 https://www.bbc.co.uk 的情况下你会被重定向到 http://www.bbc.co.uk 所以这就是为什么你看到额外的响应和 0 为 len(res.text) 返回的原因，你可以看到下面的输出：

In [11]: def cb(res, **kwargs):
   ....:         print(res.status_code)
   ....:         print("**** Response", time.time(), len(res.text))
   ....:     

In [12]: for i in range(10):
   ....:         unsent = grequests.get("https://www.bbc.co.uk", hooks={'response': cb})
   ....:         print("Request", time.time())
   ....:         grequests.send(unsent, grequests.Pool(1))
   ....:         gevent.sleep(0.2)
   ....: gevent.sleep(5)
   ....: 
('Request', 1459368704.32843)
301
('**** Response', 1459368704.616453, 0)
('Request', 1459368704.618786)
301
('**** Response', 1459368704.937159, 0)
('Request', 1459368704.941069)
200
('**** Response', 1459368704.943034, 141486)
301
('**** Response', 1459368705.496423, 0)
('Request', 1459368705.498991)
200
('**** Response', 1459368705.50162, 141448)
301
('**** Response', 1459368705.784145, 0)
('Request', 1459368705.785769)
200
('**** Response', 1459368705.786772, 141486)
301
('**** Response', 1459368706.110865, 0)
('Request', 1459368706.114921)
200
('**** Response', 1459368706.116124, 141448)
301
('**** Response', 1459368706.396807, 0)
('Request', 1459368706.400795)
200
301
('**** Response', 1459368706.756861, 0)
('Request', 1459368706.76069)
200
('**** Response', 1459368706.763268, 141448)
('**** Response', 1459368706.488708, 141448)
301
('**** Response', 1459368707.065011, 0)
('Request', 1459368707.069128)
200
('**** Response', 1459368707.071981, 141448)
301
('**** Response', 1459368707.366737, 0)
('Request', 1459368707.370713)
200
('**** Response', 1459368707.373597, 141448)
301
('**** Response', 1459368707.73689, 0)
200
('**** Response', 1459368707.743815, 141448)
200
('**** Response', 1459368707.902499, 141448)

如果我们运行使用通过 https 提供服务的站点的相同代码，https://www.google.ie/ 在本例中：

In [14]: for i in range(10):
   ....:         unsent = grequests.get("https://www.google.ie/", hooks={'response': cb})
   ....:         print("Request", time.time())
   ....:         grequests.send(unsent, grequests.Pool(1))
   ....:         gevent.sleep(0.2)
   ....: gevent.sleep(5)
   ....: 
('Request', 1459368895.525717)
200
('**** Response', 1459368895.838453, 19682)
('Request', 1459368895.884151)
200
('**** Response', 1459368896.168789, 19650)
('Request', 1459368896.22553)
200
('**** Response', 1459368896.491304, 19632)
('Request', 1459368896.542206)
200
('**** Response', 1459368896.808875, 19650)
('Request', 1459368896.850575)
200
('**** Response', 1459368897.144725, 19705)
('Request', 1459368897.173744)
200
('**** Response', 1459368897.45713, 19649)
('Request', 1459368897.491821)
200
('**** Response', 1459368897.761675, 19657)
('Request', 1459368897.792373)
200
('**** Response', 1459368898.331791, 19683)
('Request', 1459368898.350483)
200
('**** Response', 1459368898.836108, 19713)
('Request', 1459368898.855729)
200
('**** Response', 1459368899.148171, 19666)

您会发现行为有所不同。我们收到 10 条回复，但没有 0 长度的回复。你应该检查你的函数中的 status_code 来验证你得到了你想要的。上面的示例解释了您在 bbc 网站上看到的内容以及其他网站很可能发生的情况。

Answer 2

grequests 的当前迭代包含以下内容：

from gevent import monkey as curious_george
curious_george.patch_all(thread=False, select=False)

有问题的部分是 select=False - 删除它或手动调用 monkey.patch_select() 可以解决问题。我不确定这是否有其他副作用。

gevent/grequests 通过 HTTPS 的奇怪阻塞行为

Strange blocking behaviour with gevent/grequests over HTTPS

python

https

gevent

greenlets

grequests