Couchbase SDKs can connect to any node in a cluster; at runtime SDKs also automatically receive information from Couchbase Server if any nodes are unavailable. If a node that is used by your application fails, the SDK will be informed by Couchbase Server and mark that node as down and will also have information about alternate nodes that are still available. You use the Couchbase Admin tool to manually indicate a node has failed, or you can configure couchbase Server to use auto-failover. For more information, see Couchbase Server 1.8 Manual.
During node failure, Couchbase SDKs will get errors trying to read or write any data that is on a failed node. Couchbase SDKs are still able to read and write to all other functioning nodes in the cluster. After the node failure has been detected and the node has been failed-over, SDKs will be updated by the Couchbase Server and will resume functioning with the cluster and nodes as they normally would. In this way, Couchbase SDKs and the applications you build on them are able to cope with transient node failures and still conduct reads and writes.
For more information about node failover, see Couchbase Server Manual 2.0, Node Failover.