The surefire default setting for test order is "filesystem" which executes tests as they are listed when traversing the filesystem. Without knowing OS specifics, this order is arbitrary and hard to reproduce locally. Should either use truly random ordering (to improve detecting rate of test failures due to ordering) or should use a truly determined ordering (to improve reproducability)

Pull request for solution 2: https://github.com/camunda/camunda-bpm-platform/pull/226