Currently, the batch subsystem behaves like these:

• If a job is {}not stopped{} during a "server crash", then the job will always be kept in STARTED status.
• If a job is {}stopped{} during a "server crash", then the job will be moved into UNKNOWN status.
• If a {}crashed{} job stays in STARTED status and is stopped after a "server restart", the job will be moved into UNKNOWN status.
• The UNKNOWN job can’t be restarted manually in the Admin Console. (I haven't checked the CLI side)

The current implementation could be improved in these places:

• No matter whether a job is {}stopped{} or not during a "server crash", the job {}shouldn’t{} be kept in UNKNOWN status.
• The STARTED job should be able to be stopped correctly, and it {}shouldn’t{} be moved to UNKNOWN status after manually stopping it.