I've been using SparkConf on my project for quite some time now to store configuration information for its various components. This has worked very well thus far in situations where I have control over the creation of the SparkContext & the SparkConf.
I have run into a bit of a problem trying to integrate this same approach to the use of the shell, however. I have a bunch of properties in a properties file that are shared across several different types of applications (web containers, etc...) but the SparkConf ignores these properties because they aren't prefixed with spark.* Is this really necessary? It's not really stopping people from adding their own properties and it limits the power of being able to utilize one central configuration object.