CPU task on GPU cluster

Organization Name:
Advantech

Short summary about the issue/question:
CPU task behavior on GPU cluster

Brief what process you are following:
Did some test on GPU cluster capability

SKU setup in cluster: 1GPU, 3CPU, 90G RAM

1. Task with 1GPU, 2CPU and 3G RAM: Running
2. Task with 1CPU and 3G RAM: Running
3. Task with 2CPU and 4G RAM: Waiting . Log information is listed below:

I1215 02:42:59.219909       6 scheduling_queue.go:908] About to try and schedule pod default/5f7c9b16f10dc398a943e8aea61d534b-jupyter-0
I1215 02:42:59.219969       6 scheduler.go:453] Attempting to schedule pod: default/5f7c9b16f10dc398a943e8aea61d534b-jupyter-0
I1215 02:42:59.272955       6 factory.go:647] Unable to schedule default/5f7c9b16f10dc398a943e8aea61d534b-jupyter-0: no fit: 0/5 nodes are available: 1 Pod is waiting for preemptible or free resource to appear: insufficient capacity in VC devgpu, 1 node(s) didn't match node selector.; waiting
I1215 02:42:59.273003       6 factory.go:742] Updating pod condition for default/5f7c9b16f10dc398a943e8aea61d534b-jupyter-0 to (PodScheduled==False, Reason=Unschedulable)
I1215 02:42:59.275557       6 generic_scheduler.go:1118] Node testcpu0-1111972-iaas is a potential node for preemption.
I1215 02:42:59.275574       6 generic_scheduler.go:1118] Node testgpu0-1111945-iaas is a potential node for preemption.
I1215 02:42:59.275577       6 generic_scheduler.go:1118] Node testgpu1-1112050-iaas is a potential node for preemption.
I1215 02:42:59.275580       6 generic_scheduler.go:1118] Node testcpu1-1111975-iaas is a potential node for preemption.

Why the different behavior for dry-run No. 2 and No.3 ?

How to reproduce it:
Assign task with 0 GPU and multiple CPU to GPU cluster.

OpenPAI Environment:

• OpenPAI version:
  v1.0.1
• Cloud provider or hardware configuration:
  OpenStack VM
• OS (e.g. from /etc/os-release):
  ubuntu 16.04.6