MapReduce job works fine when the corresponding mapper and reducer tranformations do not contain sub tranformations. But if we include sub transformations in the Mapper tranformation, Mapreduce job fails in Hadoop with error message "Failed to initialize step". This is happening as the Hadoop job is unable to reference the Pentaho repository for sub tranformation.
Workaround we have is to use VFS to save sub tranformation KTRs in HDFS and then refer to the HDFS location from Mapper tranformation. This is getting cumbersome as any change in the sub ktr now has to be saved in repository first and a copy of it transferred to HDFS.
Please consider fixing this issue so that sub transformations can be referenced from repository itself.