With the memcached protocol level flush_all command disabled as of 1.8.1, we need a feature to achieve a similar kind of flush.
From earlier discussions, this would be available through the REST interface (disabled by default, with some bucket-level configuration option to enable). In previous discussions, we said it should similar semantics to flush_all from a client. By this I mean a request to REST to 'flush the bucket' would not remove the definition of the bucket like a delete, and clients would not be dropped or receive any errors while the cluster does whatever needs to be done to remove the data from the cluster.
The full specification is TBD.