Unfortunately, I am not sure of a simple way to provide the partitioner information with the existing implementation. Currently the KuduRDD does not override the RDD partitioner <https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/rdd/RDD.scala#L141>, though it probably could as an improvement.
Would you like to file a Kudu jira to track the work? Would you be interested in contributing the improvement? I am curious to know, how are you planning to use the knowledge of the original Kudu partitioning and how is it useful to your Spark workflow? Thanks, Grant On Wed, Nov 14, 2018 at 2:41 AM Dmitry Pavlov <[email protected]> wrote: > Hi guys > > I have a question about Kudu with Spark. > > For example there is a table in kudu with field record_id and following > partitioning: > HASH (record_id) PARTITIONS N > > Is it possible to load records from such table in key value fashion with > correct partitioner information in RDD? For example RDD[(record_id, row)] > Because when i try to use kuduRDD in spark the partitioner has None value > so im losing information about original (kudu) partitioning. > > Thanks -- Grant Henke Software Engineer | Cloudera [email protected] | twitter.com/gchenke | linkedin.com/in/granthenke
