After a server description is invalidated (typically from a socket hangup during an application operation), its RTT should be reset. The next time we successfully connect to it, the new RTT begins a new series of moving averages.

The Server Discovery And Monitoring Spec says about a server description, "if it changes from a known type to Unknown its RTT is set to null."

The C Driver instead records an RTT of 0 and factors this into the moving average. Therefore, a server that we have failed to reach for the last few checks and then successfully connected to, appears to have a very low latency and it is preferred for reads.