[YARN-5128] Investigate potential race condition in the scheduler nodeUpdate() method

• Type: Task
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Assignee: Wilfred Spiegelenburg
• Reporter: Ray Chiang
• Created: 23/May/16 20:03
• Updated: 26/Feb/20 05:29
• Resolved: 29/Mar/18 06:20

Description

This section of code exists in the various schedulers in the method nodeUpdate():

    // If the node is decommissioning, send an update to have the total
    // resource equal to the used resource, so no available resource to
    // schedule.
    // TODO: Fix possible race-condition when request comes in before
    // update is propagated
    if (nm.getState() == NodeState.DECOMMISSIONING) {
      this.rmContext
          .getDispatcher()
          .getEventHandler()
          .handle(
              new RMNodeResourceUpdateEvent(nm.getNodeID(), ResourceOption
                  .newInstance(getSchedulerNode(nm.getNodeID())
                      .getAllocatedResource(), 0)));
    }

Investigate the TODO section.

Issue Links

• duplicates: YARN-4677 RMNodeResourceUpdateEvent update from scheduler can lead to race condition
• is related to: YARN-5047 Refactor nodeUpdate across schedulers