Checkpoints record what has been processed for a specific query, and as such only need to be defined when writing (which is how you "start" a query).
You can use the DataFrame created with readStream to start multiple queries, so it wouldn't really make sense to have a single checkpoint there. On Mon, Sep 11, 2017 at 2:36 AM, kant kodali <kanth...@gmail.com> wrote: > Hi All, > > I was wondering if we need to checkpoint both read and write streams when > reading from Kafka and inserting into a target store? > > for example > > sparkSession.readStream().option("checkpointLocation", "hdfsPath").load() > > vs > > dataSet.writeStream().option("checkpointLocation", "hdfsPath") > > Thanks! >