We have seen LSAN report leaks of a 28-byte struct sockaddr_in6 originating from a malloc in libresolv's __res_context_send. Investigation showed that there are living pointers to the object in the TLS of still-running ASIO-spawned threads when the leak-detector is running. We have seen issues in the past where LSAN reports leaks of data from still-running threads, particularly with thread_local data.
Normally we try and join the offending thread(s) but since this is an ASIO-spawned one and it's not obvious what NetworkInterface(s) it originates from, I think we're better off just suppressing it until we have the shutdown path "clean" and ensure we join all threads before shutdown, particularly since this leak has an identifiable signature outside of mongo code.