Klever

Evgeny Novikov wrote:

It looks strange that code for processing solved and failed jobs in service.utils.FinishJobDecision#_get_status_ differs too much. In particular, when Core doesn't send its finish report and fails, Bridge decides to set status "Terminated" intended for reporting critical issues with schedulers.

If Core doesn't send finish report it is NOT properly failed Job. "Failed" status is intended only for jobs with unknown report for Core. So either "Terminated" or "Corrupted" can be set for jobs without finish Core report.

Status that function service.utils.FinishJobDecision#_get_status_ get in args is status set by scheduler, but then Bridge check everything and set proper status.

Vladimir Gratinskiy wrote:

Evgeny Novikov wrote:

It looks strange that code for processing solved and failed jobs in service.utils.FinishJobDecision#_get_status_ differs too much. In particular, when Core doesn't send its finish report and fails, Bridge decides to set status "Terminated" intended for reporting critical issues with schedulers.

If Core doesn't send finish report it is NOT properly failed Job. "Failed" status is intended only for jobs with unknown report for Core. So either "Terminated" or "Corrupted" can be set for jobs without finish Core report.

"Corrupted" should be used since I described above when "Terminated" can be used.

Status that function service.utils.FinishJobDecision#_get_status_ get in args is status set by scheduler, but then Bridge check everything and set proper status.

Bridge always will be a final arbiter. Moreover, it knows more than schedulers since it processes reports in addition.