The batch spec group is pondering getting rid of global transaction around partition analyzers, or something similar, in the next spec version.
See spec bug for detailed discussion:
PartitionAnalyzer transaction boundary on partitioned chunk (and batchlet) step need clarification
A related JBeret issue:
Exception if transaction timeout occured before partitions complete
These spec changes, when implemented in JBeret, will not be backward compatible with existing JBeret and WildFly versions. So I'd like check with you, JBeret and WildFly users and batch application developers, how this change will affect your app. Feel free to add your comment to the above batch spec bug, or here.
Thanks in advance.