Users reasonably expect max_pool_size to enforce the maximum number of concurrent connections. In fact, max_pool_size is the max number of idle sockets that threads / greenlets can return to the pool before the pool starts discarding returned sockets. PyMongo does not track the number of concurrent connections in use, nor enforce a limit.

Follow the lead of other drivers (Java?) and enforce a maximum number of concurrent connections with a semaphore. Leave unbounded by default until we're ready to make a breaking change. Consider renaming max_pool_size to something more descriptive.