When a job execution is killed or server crashed, the batch runtime does not have enough time to update its batch status to FAILED in job repository, so its status is left as STARTING or STARTED or STOPPING. When user tries to restart the failed job execution, the batch runtime will not restart it, because only FAILED or STOPPED job execution can be restarted.

jberet-core should be able to detect such failed job executions, update its status in job repository, and properly restart them.