In the command line help message, it says
--input,-I:String BAM/SAM/CRAM file containing reads This argument must be specified at least once.
However, if we actually give multiple input files, it says
org.broadinstitute.hellbender.exceptions.UserException: Sorry, we only support a single reads input for spark tools for now.
On the other hand, if we specify the input parameter as the folder containing all partial bam files, it actually works. Could you explain how this feature works now? We are using GATK 4 master branch, commit b82b5b6c5cbef8973b373edfb314cf42bea5eb1a, with Spark 2.0.2.