@nenzax -
In this case, the operation will fail-fast; however, the SDK will try to reconnect with node every 1000ms by sending noop’s until it connects. The reason it does this is to not spend time sending operations to a node that is offline until it has gone back online.
That was the theory, in-practice TCP resets due to flakey networks and net-appliances (firewalls, LB’s, etc) can happen frequently creating a “false” positive. So, the behavior was changed in 2.2.X so that a certain number of failures will have to happen within a timespan before the node is considered offline or unreachable from the client (this doesn’t actually mean the physical node is down, it just means the client cannot connect).
-Jeff