Step never launched when restarting a failed loop flow job [BATCH-2703]

Damien Lenci opened BATCH-2703 and commented

My job is using a flow of 3 steps (A, B, C) in a loop.
When no error occurs, the job executes fine : A -> B -> C -> A -> B -> C ->...
If an error occurs on step B and I try to restart the job, I would expect the following execution :
B -> C -> A -> B -> C ->...
Instead, step A is never executed again, resulting in the execution B -> C -> B -> C -> ...

Looking at SimpleStepHandler, maybe the following behaviour shoud be extended to remove the last execution of an already skipped step :

if (stepExecutionPartOfExistingJobExecution(execution, lastStepExecution)) {
	// If the last execution of this step was in the same job, it's
	// probably intentional so we want to run it again...
	logger.info(String.format("Duplicate step [%s] detected in execution of job=[%s]. "
					+ "If either step fails, both will be executed again on restart.", step.getName(), jobInstance
					.getJobName()));
	lastStepExecution = null;
}

Affects: 3.0.7

Issue Links:

• BATCH-2533 Restarting a job that failed before going back to a previous step is not working well

Start by reading SimpleStepHandler and the linked BATCH-2533 context. Trace how the failed flow job selects the next step on restart, especially when a previously skipped step is encountered. Done means a restart follows B -> C -> A rather than repeatedly skipping A, while preserving the expected looping behavior.