Adam Kawa created MAPREDUCE-5660:
------------------------------------

             Summary: Log info about possible thrashing (when using 
memory-based scheduling in Capacity Scheduler) is not printed
                 Key: MAPREDUCE-5660
                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-5660
             Project: Hadoop Map/Reduce
          Issue Type: Bug
          Components: capacity-sched, mrv1, tasktracker
    Affects Versions: 1.2.1
            Reporter: Adam Kawa
            Priority: Trivial


There is a tiny, but confusing when troubleshooting, bug in TaskTracker code:
{code}
if (totalMemoryAllottedForTasks > totalPhysicalMemoryOnTT) {
  LOG.info("totalMemoryAllottedForTasks > totalPhysicalMemoryOnTT."
      + " Thrashing might happen.");
} else if (totalMemoryAllottedForTasks > totalVirtualMemoryOnTT) {
  LOG.info("totalMemoryAllottedForTasks > totalVirtualMemoryOnTT."
      + " Thrashing might happen.");
}
{code}
totalMemoryAllottedForTasks is calculated in megabytes, while 
totalPhysicalMemoryOnTT (and totalVirtualMemoryOnTT) is calculated in bytes. 
totalMemoryAllottedForTasks should be converted to bytes for a correct 
comparison.



--
This message was sent by Atlassian JIRA
(v6.1#6144)

Reply via email to