As mentioned in #5576 (comment), according to automated analysis, chaining doesn't work right with the walltime requrement. Right now, I think we will happily chain to a job even if the remaining time in the allocation is not enough for its full walltime.
If we just chain to jobs that will fit in the remainder of the original job's requested walltime, automated analysis suggests that we then won't actually do a lot of chaining. If we set a --defaultWalltime, and don't mark jobs with custom walltime requirements, then all the jobs will have the same walltime requirement, and no job will ever have enough time left in its allocation after finishing to run another job.
To solve this, we might want to avoid enforcing walltime requirements on local jobs, and/or we might want to request extra time or have some notion of minimum and maximum walltime.
┆Issue is synchronized with this Jira Story
┆Issue Number: TOIL-1865
As mentioned in #5576 (comment), according to automated analysis, chaining doesn't work right with the
walltimerequrement. Right now, I think we will happily chain to a job even if the remaining time in the allocation is not enough for its fullwalltime.If we just chain to jobs that will fit in the remainder of the original job's requested
walltime, automated analysis suggests that we then won't actually do a lot of chaining. If we set a--defaultWalltime, and don't mark jobs with custom walltime requirements, then all the jobs will have the samewalltimerequirement, and no job will ever have enough time left in its allocation after finishing to run another job.To solve this, we might want to avoid enforcing
walltimerequirements on local jobs, and/or we might want to request extra time or have some notion of minimum and maximum walltime.┆Issue is synchronized with this Jira Story
┆Issue Number: TOIL-1865