[YARN-8443] Total #VCores in cluster metrics is wrong when CapacityScheduler reserved some containers

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.9.0, 3.1.0, 3.2.0
• Fix Version/s: 3.2.0, 3.1.1, 2.9.2, 3.0.4
• Component/s: webapp
• Hadoop Flags: Reviewed
• Assignee: Tao Yang
• Reporter: Tao Yang
• Created: 20/Jun/18 12:35
• Updated: 25/Jun/18 02:01
• Resolved: 25/Jun/18 01:37

Description

Cluster metrics on the web UI will give wrong Total Vcores when there is reserved container for CapacityScheduler.
Reference code:

ClusterMetricsInfo.java

    if (rs instanceof CapacityScheduler) {
      CapacityScheduler cs = (CapacityScheduler) rs;
      this.totalMB = availableMB + allocatedMB + reservedMB;
      this.totalVirtualCores =
          availableVirtualCores + allocatedVirtualCores + containersReserved;
       ...
    }

The key of this problem is the calculation of totalVirtualCores, containersReserved is the number of reserved containers, not reserved VCores. The correct calculation expression of totalVirtualCores should be (availableVirtualCores + allocatedVirtualCores + reservedVirtualCores)