Hey @hubo3085632,
Auto-failover is a hard failover. More information on hard failover’s can be found here:
http://docs.couchbase.com/admin/admin/Tasks/hard-failover.html
In terms of what happens to data in the unfortunate event that has not yet been replicated to other nodes, the data will be lost. However, losing a small chunk of data on one node is much better than corrupting or losing all your data because of having the bad node staying connected in your cluster.
There are things put into place to prevent false positives, so you can feel safe in knowing that your nodes are not just going to start doing auto-failovers.
Let me know if you have other questions.
Best,