In Couchbase, failing over a server (and thus, transferring load to replica servers for the document IDs that were previously "mastered" on that server) can be done either manually, automatically, or programmatically.

Failover within Couchbase is the process that immediately activates any replica vBuckets for the vBuckets that were active on the node being failed over.

Ideally, you should only fail over a node that is down and also only if you know you have replicas available elsewhere in the cluster:

If a node is not down, you should attempt to use the remove/rebalance functionality as this will ensure all the data is removed, re-replicated and maintained safe. See Section 5.3.3, “Rebalancing”. Failing over a live node may introduce a small data-loss window as any data "in flight" (meaning not yet replicated) will be lost when the failover takes place. If you don't have available replicas, failing over the node will result in perceived data loss as those vBuckets will be activated as empty elsewhere in the cluster. You can still recover the data, but it will not be immediately available.