Starting in version 1.2, the driver discovers and monitors the server topology using an event loop with a non-blocking socket per server, to check all servers in parallel.

If one of those checks times out or the connection is refused, the driver does not record the error.

The Server Discovery And Monitoring spec suggests recording each server check's error and summing all errors to report why no server is available:

https://github.com/mongodb/specifications/blob/master/source/server-discovery-and-monitoring/server-discovery-and-monitoring.rst#network-error-when-calling-ismaster