Process only one task of jobs in reclaim action

[zionwu]

I am using reclaim action and notice that it will only process only one task of jobs:
https://github.com/volcano-sh/volcano/blob/master/pkg/scheduler/actions/reclaim/reclaim.go#L107

             // Found "high" priority job
		if jobs, found := preemptorsMap[queue.UID]; !found || jobs.Empty() {
			continue
		} else {
			job = jobs.Pop().(*api.JobInfo)
		}

		// Found "high" priority task to reclaim others
		if tasks, found := preemptorTasks[job.UID]; !found || tasks.Empty() {
			continue
		} else {
			task = tasks.Pop().(*api.TaskInfo)
		}

The above code is to find one task from the job and use it to reclaim resources. However, the job is not pushed back to jobs(the priority queue) so it will not be processed again.

I wonder if this behavior is by design? If so, why?
Or is it a bug ? We should push the job back to jobs at this line https://github.com/volcano-sh/volcano/blob/master/pkg/scheduler/actions/reclaim/reclaim.go#L198

[k82cn]

I think we should put task back to the queue :)

btw, what's your use case for reclaim?

[k82cn]

/area scheduling
/kind bug
/priority important-soon

[zionwu]

@k82cn, thank you for the quick reply.
My use case is as follow:
we have two queues, one for high priority jobs like production training job, another for low priority jobs like testing training job. If the queue for production is running out of resource, we want to reclaim resource from the queue for testing.

Another question about reclaim is why we push the queue back to queues only when assigned is true? I think we should always push the queue back:
https://github.com/volcano-sh/volcano/blob/master/pkg/scheduler/actions/reclaim/reclaim.go#L197

                if assigned {
			queues.Push(queue)
		}

[k82cn]

Another question about reclaim is why we push the queue back to queues only when assigned is true?

If we did not assign any resource in this loop, that means there's no resource for it; so will not consider it in next loop.

[zionwu]

Another question about reclaim is why we push the queue back to queues only when assigned is true?

If we did not assign any resource in this loop, that means there's no resource for it; so will not consider it in next loop.

If assigned is false, it means there is no resource for the task of the job, but it doesn't mean there is no resource for the task of other job in the same queue.
For example, if queue A have job1 and job2. If task of job1 failed to reclaim, we should not consider job1 again, but we should consider job2. Therefore we need to put queue A back to the priority queue.

Does it make sense to you?

[k82cn]

yes, that makes more sense :)

[k82cn]

fixed by #631