I was just recently reading the code and noticed this problem. We do delete "Port" objects when they're not used, but when two Ports share a socket, it's the  BasicPortAllocatorSession that owns the socket: https://codesearch.chromium.org/chromium/src/third_party/webrtc/p2p/client/basicportallocator.h?q=basicportall&sq=package:chromium&l=368

We need to either change this model, or start doing some kind of reference counting.

Any chance this long-standing issue can be addressed soon?

A seemingly not-so-hard-to-implement fix could save gazillions of kWh for computers all over the world prevented by Chromium/Chrome from suspending on inactivity.

It's not the highest priority right now, but if it's impacting you, you're welcome to take a stab at it. https://webrtc.org/contributing/

Argh, I moved off Chromium in 2015, but chances are I'll take a look at it.

Is this bug fixed? PeerConnection doesn't close UDP ports after calling peerconnection close, And this maybe cause too many open files for Android?

The bug is not fixed, but calling PeerConnection::Close *will* close all ports.

We call PeerConnection::Close， but stun binding request send periodically， and do not know when udp port are going to be closed

I find the bug is fixed, after calling PeerConnection::Close, then PeerConnection::~PeerConnection is called， and udp port are going to be closed.

Previous code has bug, because stl container hold PeerConnection for a long time.