I've put up a short video that shows a client session connected to a RAC SCAN IP address failing over to the surviving RAC instance when the database node goes down
Video demonstration of RAC Client Query Session and Query Failover
When a client connects to a RAC Instance via a SCAN listener : If the node (Linux in this case) goes down abnormally, Grid Infrastructure does a failover of the SCAN IP address to any of the surviving nodes. In this demo, I have two connections to two different SCAN IPs, one on node1 and the other on node2 (i.e. two different Linux DB Servers running RAC). When node2 goes down abruptly, the user's session (using SQL*Developer in this case) will automatically reconnect when the SCAN IP is migrated to the surviving node (node1) and can re-query.
By default, Transactions do not failover (as I show from the alert log, the node1 instance does a recovery and rollback of failed transactions of node2). However, you CAN build Application Continuity into your architecture to failover and resume transactions as well !!