It seems that there is a race condition in the XsltAggregatorStrategy. The following code:
means that potentially one thread might be still working on the initialize call while another thread passes through the gate of the if statement and executes the xslt.process() call. If the initialization is not quite complete, the XsltBuilder will throw an exception complaining that the template has not been set.
I have written a small Spring Boot project (which I am attaching to this Jira issue) that demonstrates this. I understand that this only happens when parallelization is present during the xslt member initialization above, and that it sorts itself out in subsequent calls, but I have managed to see this during testing, so it is not that impossible for it to happen.