[YARN-6158] FairScheduler: app usage can go to negative

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 3.0.0-alpha2
• Component/s: fairscheduler, resourcemanager
• Target Version/s: 3.0.0-alpha4
• Assignee: Unassigned
• Reporter: Miklos Szegedi
• Created: 08/Feb/17 02:19
• Updated: 17/Feb/17 22:08
• Resolved: 17/Feb/17 22:08

Description

FiCaSchedulerApp.containerCompleted checks, if the container being completed is in the active list:

      // Remove from the list of containers
      if (null == liveContainers.remove(containerId)) {
        return false;
      }

Fair scheduler should do the same, otherwise multiple different container close events leave the application with negative resource usage in queue.getMetrics().releaseResources and attemptResourceUsage.decUsed

Issue Links

• duplicates: YARN-3933 FairScheduler: Multiple calls to completedContainer are not safe