Python socket connection pool resilience under rotating cellular proxy gateways & CGNAT changes

[pxm2net]

Hi everyone,

I wanted to share a small scripting pattern and ask how others handle upstream socket resets when running automated batch telemetry and headless browser tasks (Playwright/Selenium) over rotating 4G/5G mobile gateways on Debian/Siduction.

When cascading requests through cellular proxy endpoints, mobile network operators periodically rotate bearer IPs behind Carrier-Grade NAT (CGNAT). Under IETF RFC 7230 Section 6.3 persistent connection standard, clients expect persistent TCP connections, but when the carrier IP renews, active sockets in the local connection pool receive unexpected

Code Select Expand

ConnectionResetError: [Errno 104] Connection reset by peer rather than an orderly termination.

To avoid failing long-running jobs or crashing browser contexts, we implemented an explicit backoff and socket recycling wrapper using urllib3 and requests adapters:

Code Select Expand

import time
import requests
from requests.adapters import HTTPAdapter
from urllib3.util.retry import Retry

class ResilientProxySession(requests.Session):
    def __init__(self, proxy_url, max_retries=3):
        super().__init__()
        self.proxies = {'http': proxy_url, 'https': proxy_url}
        retries = Retry(
            total=max_retries,
            backoff_factor=1.5,
            status_forcelist=[502, 503, 504],
            raise_on_status=False
        )
        adapter = HTTPAdapter(max_retries=retries, pool_connections=10, pool_maxsize=10)
        self.mount('http://', adapter)
        self.mount('https://', adapter)

For headless browser orchestration, similar connection handling and proxy server routing strategies are outlined in the Playwright Python network proxy documentation and comparative benchmarks are detailed in this Selenium and Playwright mobile proxy automation architecture guide. Standard Python socket options can also be referenced in the Python urllib socket documentation.

Do you typically wrap socket renewal at the application layer with backoff adapters like this, or do you handle gateway renegotiations via local systemd watchdog / redsocks routing tables?

Curious to hear what patterns people find most reliable for long-running network workers on Siduction.