[YARN-7527] Over-allocate node resource in async-scheduling mode of CapacityScheduler

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.0.0-alpha4, 2.9.1
• Fix Version/s: 3.1.0, 2.10.0, 3.0.3
• Component/s: capacityscheduler
• Hadoop Flags: Reviewed
• Assignee: Tao Yang
• Reporter: Tao Yang
• Created: 17/Nov/17 08:13
• Updated: 12/Apr/18 02:21
• Resolved: 12/Apr/18 02:21

Description

Currently in async-scheduling mode of CapacityScheduler, node resource may be over-allocated since node resource check is ignored.
FiCaSchedulerApp#commonCheckContainerAllocation will check whether this node have enough available resource for this proposal and return check result (ture/false), but this result is ignored in CapacityScheduler#accept as below.

commonCheckContainerAllocation(allocation, schedulerContainer);

If FiCaSchedulerApp#commonCheckContainerAllocation returns false, CapacityScheduler#accept should also return false as below:

if (!commonCheckContainerAllocation(allocation, schedulerContainer)) {
  return false;
}