Custom network overlay does not work for all ip addresses

[n9iels]

General

After having troubles with following the "Creating your first overlay" tutorial (see: #698) I noticed that my laptop was not able to find any peers on the office network, but that it was working at my home network. The main difference is that my home network gives my laptop a IP in the range of 192.168.2.x, while the office network gives my laptop a IP in the range of 10.x.x.x. I can also reproduce this issue when using a mobile hotspot which gives me a IP in t he range of 192.168.43.x.

Reproducing the issue

1. Implement the "Printing the known peers" section of the "Creating your first overlay" tutorial
2. Verify that the output is as expected (the described output in the tutorial)
3. Connect to a mobile hotspot
4. Re-run the script and verify that no peer could be found anymore

System information

• Latest version of master (f9afbb94f710717b6cf40e1a228cb3168dd2ce68)
• Python 3.7.5

[qstokkink]

Interesting, do you have netifaces installed?

[n9iels]

Yes, I used the provided requirements.txt and I verified that the corresponding methods are used. For example in endpoint.py the get_lan_address_without_netifaces() method is not used.

[qstokkink]

Reproducing this will probably take a little while, sounds like an issue with the type of NAT (though it really should work). At some point we'll have to patch up and systematically test configurations with our NAT setup. I'll try to reproduce your setup and hope the problem occurs.

@synctext official request for another student on NAT traversal 😃

[qstokkink]

I actually managed to reproduce this today. For some reason the WAN->LAN address translation/estimation is broken when two machines are behind the same WAN address.

EDIT: nevermind, I was running an experiment that changed the walking.

[qstokkink]

This should have been fixed by #1133, at least to the point that it can reasonably be fixed by IPv8's NAT puncturing approach. Different approaches than the current one are discussed in Tribler/tribler#2754 (superseding this issue).