with default hadoop the com.sun.xml.* implementation is used which is compatible with spark-xml. with later hadoop versions the default is com.ctc.wstx.stax.* which will not work with spark-xml. this will force correct version