I just used spark streaming. I have a few questions about checkpoint:

There are two types of
1. checkpoint, one for meta for driver and one for data. The manual says that the checkpoint of data will be written only if you use stateful transformation. So, will I write if I don"t use stateful transformation,data "s checkpoint data? If I don"t write, where can I get the lost rdd when I re-restart?
The checkpoint in
2. batch job I can specify which rdd to write, or at which step to write checkpoint. So in streaming, do I need to use foreachRDD {rdd.checkpoint ()}?
3. if I don"t show write rdd.checkpoint () in spark streaming, how does spark decide which rdd should write the data checkpoint file?