What is the best / most appropriate way to configure your own implementation of the StepExecutionSplitter interface? At the moment I am providing my own version of the class at the head of the classpath, but this is not an ideal approach.

The underlying issue is that I have a restartable job that is being partitioned. Some of the partition are completing, others are failing, and on restart, only the incomplete partitions are being re-run. When the job finally succeeds, the only contexts that are passed to the StepExecutionAggregator as those that were a part of the latest execution of the job, resulting in earlier completed steps (from previous executions) not being a part of the final aggregated results.

Unfortunately the allowStartIfComplete property does not help in this case, as the a fresh context will be used.

Cheers,
-Daniel