What happened + What you expected to happen

If jobs are submitted with the same ID, the second one will fail with an internal unfriendly error (though it hints at the root cause). Even if the first job would have succeeded, its status is overwritten with the failed status from the second job.

❯ ray job submit --submission-id blah3 --no-wait -- echo hi again & ray job submit --submission-id blah3 --no-wait -- echo hi again

❯     ray job status blah3
Job submission server address: http://127.0.0.1:8265

------------------
Job 'blah3' failed
------------------

Status message: Failed to start Job Supervisor actor: The name _ray_internal_job_actor_blah3 (namespace=SUPERVISOR_ACTOR_RAY_NAMESPACE) is already taken. Please use a different name or get the existing actor using ray.get_actor('_ray_internal_job_actor_blah3', namespace='SUPERVISOR_ACTOR_RAY_NAMESPACE').

I would expect that (1) the first command should succeed and its status should reflect that, and (2) the second should fail with RuntimeError: Job blah3 already exists.. This currently happens if the first command is given a second or so to run and update its internal JobInfo, but this should still happen even if the commands are issued right after one another.

Versions / Dependencies

master, MacOS, Python 3.8

Reproduction script

Above

Issue Severity

Medium: It is a significant difficulty but I can work around it.