The problem is described in details in JBTM-751, which also describes a fix by ignore solution.

The bug is already fixed for IronJacamar by JBJCA-599, revision 111645
That solution is much nicer, but I can't judge if it's backwards compatible.

I tested my "fix by ignore"-solution for a day in an environment that continuous to throw the mysql deadlock, and all processes and transactions continued to work correctly.