Run Spark Shell with a Different User Directory

I needed to be able to the run Spark’s shell using a different home directory due to a permissions issue with my normal HDFS home directory. A little known feature with spark is that it allows overriding Hadoop configuration properties!


spark-shell --master yarn --conf spark.hadoop.dfs.user.home.dir.prefix=/tmp/user

Normal Spark configuration parameters can be provided using --conf but you can also override Hadoop configurations as well by prefixing the normal Hadoop property name with spark.hadoop.

In this case, the Hadoop property is dfs.user.home.dir.prefix which by default is set to /user. By passing spark.hadoop.dfs.user.home.dir.prefix that value becomes /tmp/user. Now when I start up Spark shell a directory is created at /tmp/user/[my username]!