How distributed countByKey works in KStream ?

[Tommy Go]

Hi,

For "word count" example in Hadoop, there are shuffle-sort-and-reduce
phases that handles outputs from different mappers, how does it work in
KStream ?