While testing cluster failover, we’ve noticed that the client looses connections for the 7 or 8 seconds it takes to fail over. Is there a way to make the failover more transparent?
For example, when a user is copying a file, the copying fails. The user has to recopy the file after connection is reestablished.