I found that every time the connection is renegotiated on a bundle connection, new ice candidates are generated for that mline. It also seems candidates that are unused/unreachable continue to keep sockets open even after the connection has established.
The result is a hundreds to thousands of lingering open sockets through long sessions with lots of renegotiations (my use case).
I believe fixing the latter, closing/ending unused candidates after the ice transport connects, will fix both bugs.
Related to my renegotiation work:
I found that every time the connection is renegotiated on a bundle connection, new ice candidates are generated for that mline. It also seems candidates that are unused/unreachable continue to keep sockets open even after the connection has established.
The result is a hundreds to thousands of lingering open sockets through long sessions with lots of renegotiations (my use case).
I believe fixing the latter, closing/ending unused candidates after the ice transport connects, will fix both bugs.