The root of the race condition that we are encountering is, as we know, the fact that the network object that we are working with is being shared between different threads, which are mutating and reading the data from the data structure simultaneously. Specifically, the second thread in our program was mutating the data (by calling the refresh_primary() method), while the first thread was reading from the same data.
Obviously, we can simply apply locking as the synchronization mechanism for this data structure. However, we know that the tasks of acquiring and releasing locks involve a slight cost that will become substantial as the data structure is widely used across a system. As popular websites and systems (namely, MongoDB) ...
