Spark stratified sampling in dataset

Question

I have a spark dataset of 700 million rows which are partitioned by id.

Each row Is id and a value. I want to create a stratifi