milenkovicm commented on PR #2631: URL: https://github.com/apache/datafusion-ballista/pull/2631#issuecomment-6302846306
I have one philosophical question, for it let's assume we keep default timeout. From my experiences we have two types of jobs, short ones, and the ones taking forever. Short ones if scheduler gets killed it's not big deal, they are cheap to re run. Taking forever jobs are not that cheap to retry but they are going to be killed anyway as timeout is set. So instead of "waiting for a job to finish" timeout would it make sense to have "scheduler save state" timeout? If HA solution is needed we could have some kind of scheduler gateway where clients would be redirected to next scheduler which can finish the job, rehydrating the job state saved by the killed scheduler Would it make sense to take "Let it crash" approach, where we don't guarantee waiting for jobs to finish? Don't get me wrong @villebro I think this is valuable PR but I think it make sense opening this discussion -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
