为何Python的requests.get请求调用出现挂起问题？

本地运行Python requests脚本发起GET请求无响应，但curl和同一硬件的WSL环境下运行脚本正常：

复现脚本 #

基础版本：

import requests
print(requests.get('https://api.openstreetmap.org/api/0.6/node/1894790125'))

指定通用User-Agent后问题仍存在：

import requests

print(requests.get(
   'https://api.openstreetmap.org/api/0.6/node/1894790125',
    headers = { 'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/58.0.3029.110 Safari/537.36' }
))

tcpdump抓包信息 #

执行抓包命令：

sudo tcpdump -v -i eno1 'host api.openstreetmap.org'

抓包输出：

tcpdump: listening on eno1, link-type EN10MB (Ethernet), snapshot length 262144 bytes
08:17:22.463843 IP6 (flowlabel 0x69a8c, hlim 64, next-header TCP (6) payload length: 40) aaaa:bbbb:cccc:dddd:eeee:dddd:cccc:bbbb.aaaa4 > spike-06.openstreetmap.org.https: Flags [S], cksum 0xdb71 (incorrect -> 0xb5a5), seq 3910911789, win 64800, options [mss 1440,sackOK,TS val 1030348905 ecr 0,nop,wscale 7], length 0
08:17:23.482673 IP6 (flowlabel 0xd077e, hlim 64, next-header TCP (6) payload length: 40) aaaa:bbbb:cccc:dddd:eeee:dddd:cccc:bbbb.aaaa4 > spike-06.openstreetmap.org.https: Flags [S], cksum 0xdb71 (incorrect -> 0xb1aa), seq 3910911789, win 64800, options [mss 1440,sackOK,TS val 1030349924 ecr 0,nop,wscale 7], length 0
08:17:25.494643 IP6 (flowlabel 0x3cc5c, hlim 64, next-header TCP (6) payload length: 40) aaaa:bbbb:cccc:dddd:eeee:dddd:cccc:bbbb.aaaa4 > spike-06.openstreetmap.org.https: Flags [S], cksum 0xdb71 (incorrect -> 0xa9ce), seq 3910911789, win 64800, options [mss 1440,sackOK,TS val 1030351936 ecr 0,nop,wscale 7], length 0
08:17:29.687440 IP6 (flowlabel 0x7665b, hlim 64, next-header TCP (6) payload length: 40) aaaa:bbbb:cccc:dddd:eeee:dddd:cccc:bbbb.aaaa4 > spike-06.openstreetmap.org.https: Flags [S], cksum 0xdb71 (incorrect -> 0x996e), seq 3910911789, win 64800, options [mss 1440,sackOK,TS val 1030356128 ecr 0,nop,wscale 7], length 0
08:17:37.878687 IP6 (flowlabel 0xd5551, hlim 64, next-header TCP (6) payload length: 40) aaaa:bbbb:cccc:dddd:eeee:dddd:cccc:bbbb.aaaa4 > spike-06.openstreetmap.org.https: Flags [S], cksum 0xdb71 (incorrect -> 0x796e), seq 3910911789, win 64800, options [mss 1440,sackOK,TS val 1030364320 ecr 0,nop,wscale 7], length 0
08:17:54.010694 IP6 (flowlabel 0xbf2f2, hlim 64, next-header TCP (6) payload length: 40) aaaa:bbbb:cccc:dddd:eeee:dddd:cccc:bbbb.aaaa4 > spike-06.openstreetmap.org.https: Flags [S], cksum 0xdb71 (incorrect -> 0x3a6a), seq 3910911789, win 64800, options [mss 1440,sackOK,TS val 1030380452 ecr 0,nop,wscale 7], length 0
^C
6 packets captured
28 packets received by filter
3 packets dropped by kernel

关键现象 #

• 本地Python脚本持续无响应，curl访问同一URL可立即获得结果
• WSL环境下运行相同脚本，正常返回<Response [200]>
• 抓包显示仅发送TCP SYN包，无服务器的SYN-ACK回应，且所有IPv6包的校验和被标记为incorrect

核心原因：IPv6校验和异常导致连接失败 #

抓包中的cksum ... (incorrect -> ...)是关键线索：

1. 校验和不匹配：系统发送的IPv6包校验和计算异常，要么是网络栈生成包时出错，要么是网卡的校验和卸载功能故障，导致服务器丢弃这些无效包，无法建立TCP连接。
2. 协议优先差异：requests库默认优先使用IPv6连接目标地址，而curl可能自动 fallback 到IPv4；WSL环境的网络栈可能默认使用IPv4，或者其IPv6校验和功能正常，因此能正常建立连接。

解决方案 #

1. 强制requests使用IPv4
   修改脚本，强制通过IPv4发起请求：

import requests
from requests.adapters import HTTPAdapter
from urllib3.util.connection import create_connection

def force_ipv4(host, port, **kwargs):
    return create_connection((host, port), source_address=None, timeout=kwargs.get('timeout'), family=4)

session = requests.Session()
adapter = HTTPAdapter()
adapter.init_poolmanager(connection_factory=force_ipv4)
session.mount('https://', adapter)

print(session.get('https://api.openstreetmap.org/api/0.6/node/1894790125'))

2. 关闭网卡IPv6校验和卸载
   以Linux系统为例，执行命令关闭对应网卡的IPv6校验和卸载功能：

sudo ethtool -K eno1 tx-checksum-ipv6 off

测试有效后，可将该配置写入网卡永久配置文件，避免重启后失效。

3. 临时禁用系统IPv6
   若上述方法无效，可临时禁用系统IPv6，让requests自动切换到IPv4：

sudo sysctl -w net.ipv6.conf.all.disable_ipv6=1
sudo sysctl -w net.ipv6.conf.eno1.disable_ipv6=1

内容的提问来源于stack exchange，提问作者René Nyffenegger